Motion recognition system, motion recognition method, and program

Abstract

The system detects a person's state and specific actions while respecting privacy. The motion recognition system recognizes the actions of an object and includes: a storage means for storing specific action data that indicates a specific action including a predetermined series of postures and actions of at least one body part of the object; a detection means for continuously acquiring point cloud data indicating the three-dimensional coordinate position on the surface of the object and detecting the action of the object from the point cloud data; a display means for continuously displaying the action of the object detected by the detection means as a moving image from the point cloud data; an estimation means for estimating the action of the object from the continuous changes in the point cloud data; a receiving means for receiving the designation of a specific action from among a plurality of actions of the object; and a notification means for notifying that a specific action has occurred when the action of the object matches the designated specific action.

Description

Technical Field

[0001] The present invention relates to an action recognition system, an action recognition method, and a program.

Background Art

[0002] An input reception unit that reproduces and displays a first moving image of a person performing a specific action, and receives an instruction input for instructing a position related to the specific action on the frame of the reproduced and displayed first moving image during reproduction display; a second feature point group including one or more feature points included in a second moving image; and a determination unit that calculates a similarity between the first moving image and the second moving image based on a comparison result of feature amounts between the first feature point group including two or more feature points included in the first moving image, and determines whether the specific action is included in the second moving image based on the calculated similarity. When calculating the similarity, the determination unit assigns a greater weight to the comparison result of the feature amounts for the feature points closer to the instruction position by the instruction input among the first feature point group. There is known an action recognition system having the above (Patent Document 1).

[0003] A nurse call system having a nurse call sub-unit installed on a bed for a patient to call a nurse, and a nurse call main unit installed at a nurse station for responding to a call by the nurse call sub-unit, the nurse call system including a camera that images a patient on the bed from above the bed, and a state determination unit that analyzes an image captured by the camera, detects a change in the state of the patient, and outputs a first signal when a state change occurs. The nurse call main unit has a notification unit that executes a first notification operation upon receiving the first signal. The state determination unit outputs a second signal different from the first signal when detecting a predetermined action of a person including at least one of the patient and the nurse by analyzing the captured image. The camera or the nurse call main unit executes a second notification operation upon receiving the second signal. There is also known a nurse call system (Patent Document 2).

Prior Art Documents

Patent Documents

[0004] [Patent Document 1] Japanese Patent Publication No. 2022-77870 [Patent Document 2] Japanese Patent Publication No. 2023-51150 [Overview of the Initiative] [Problems that the invention aims to solve]

[0005] This invention detects a person's state and specific actions while taking privacy into consideration. [Means for solving the problem]

[0006] To solve the above problem, the motion recognition system according to claim 1 is: A motion recognition system that recognizes the actions of subjects for whom privacy protection is required, A storage means for storing specific action data that indicates a specific action including a predetermined series of postures and movements of at least one body part of the aforementioned object, By converting the light distribution pattern into a regular dot pattern with a divergence angle via a diffractive optical element, Invisible light Multiple beam-shaped Multiple projected as ranging light Array of light-emitting elements The system comprises a light source and a light receiving means for receiving reflected light reflected from the surface of the object, Without acquiring an RGB image, Point cloud data indicating the three-dimensional coordinate position on the surface of the aforementioned object. only A detection means that acquires the data in real time and detects the three-dimensional movement of the object from the point cloud data, A display means that displays the movement of the target detected by the detection means as point cloud data in real time as a moving image, Estimation means for estimating the movement of the object from the continuous changes in the point cloud data, A receiving means for receiving the designation of a specific operation among the multiple operations of the target, A notification means that notifies that the specified operation has occurred when the operation of the target is an operation that matches the specified operation, The system includes a recording means for sequentially saving the point cloud data acquired by the detection means, It is characterized by the following:

[0007] The motion recognition system according to claim 2 is the motion recognition system according to claim 1, The estimation means determines the operation of the target. To estimate, Annotation but Grant So training data Based on Learn So Using a learning model, It is characterized by the following:

[0008] The invention described in claim 3 is an action recognition system according to claim 1 or 2, The aforementioned specific action is a body movement that includes the angle and width of movement of the body part in question. It is characterized by the following:

[0010] To solve the above problem, the motion recognition method described in claim 4 is: A motion recognition method for recognizing the actions of subjects whose privacy protection is required, A storage step for storing specific action data that indicates a specific action including a predetermined series of postures and movements of at least one body part of the subject, multiple Array of light-emitting elements From the light source By converting the light distribution pattern into a regular dot pattern with a divergence angle via a diffractive optical element, Invisible light Multiple beam-shaped The step of projecting as a rangefinder light, A light receiving step of receiving reflected light reflected from the surface of the object, Without acquiring an RGB image, Point cloud data indicating the three-dimensional coordinate position on the surface of the aforementioned object. only A detection step which involves acquiring the data in real time and detecting the three-dimensional movement of the object from the point cloud data, A display step which displays the movement of the target detected in the detection step as point cloud data in real time as a moving image, An estimation step of estimating the movement of the object from the continuous changes in the point cloud data, A reception step that accepts the designation of a specific operation from among the multiple operations of the target, When the operation of the target matches the specified specific operation, a notification step of notifying that the specific operation has occurred; A recording step of sequentially storing the point cloud data acquired in the detection step, and characterized by the above.

[0011] To solve the above problems, the program according to claim 5 causes a computer to store specific operation data indicating a specific operation including a predetermined series of postures and operations of at least one body part of a target for which privacy protection is required, in a storage step; project invisible light as distance measurement light from a plurality of light sources, in a projection step; receive reflected light reflected from the surface of the target, in a light receiving step; Array of light-emitting elements from By converting the light distribution pattern into a regular dot pattern with a divergence angle via a diffractive optical element, a plurality of Multiple beam-shaped light sources acquire, in real time, point cloud data indicating three-dimensional coordinate positions on the surface of the target, and detect a three-dimensional and stereoscopic operation of the target from the point cloud data, in a detection step; Without acquiring an RGB image, display, in real time as a moving image, the operation of the target detected in the detection step as the point cloud data, in a display step; only estimate a specific operation of the target from continuous changes in the point cloud data, in an estimation step; receive a specification of the specific operation among a plurality of operations of the target, in a reception step; when the operation of the target matches the specified specific operation, notify that the specific operation has occurred, in a notification step; sequentially store the point cloud data acquired in the detection step, in a recording step; and characterized by the above.

Advantages of the Invention

[0012] only According to the invention described in claim 1, the operation of the target is detected as point cloud data with invisible and safe light onlyThe system can acquire and, while respecting privacy, detect a person's state and specific actions and behaviors, and share this information among users.

[0013] According to the invention described in claim 2, the accuracy of estimating the movement of the target can be improved.

[0014] According to the invention described in claim 3, the specific actions of the object can be visualized.

[0016] According to the inventions described in claims 4 and 5, the movement of an object is captured as point cloud data using invisible, safe light. only The system can acquire and, while respecting privacy, detect a person's state and specific actions and behaviors, and share this information among users. [Brief explanation of the drawing]

[0017] [Figure 1] This block diagram shows the functional configuration of the motion recognition system according to this embodiment. [Figure 2] This figure shows an example of an operation pattern for a specific operation in this embodiment. [Figure 3] This is a block diagram showing the functional configuration of a 3D sensor. [Figure 4] Figure 4A is a schematic diagram showing the dot pattern illumination of a 3D sensor, and Figure 4B is a diagram showing an example of the measurement range of a 3D sensor. [Figure 5] This is a schematic diagram showing an example of the arrangement of 3D sensors in a motion recognition system. [Figure 6] This flowchart illustrates the flow of the detection process for detecting the three-dimensional movement of an object in a motion recognition system. [Figure 7] This diagram conceptually illustrates an example of a process for extracting point cloud data containing only patients from acquired point cloud data. [Figure 8] This is a conceptual diagram illustrating an example of a process for removing noise from point cloud data. [Figure 9] This diagram illustrates an example of calculating a patient's 3D coordinate information using trigonometry. [Modes for carrying out the invention]

[0018] Next, specific examples of embodiments of the present invention will be described with reference to the drawings, but the present invention is not limited to the following embodiments. Please note that in the following explanation using diagrams, the diagrams are schematic, and the proportions of the dimensions may differ from those of reality. For ease of understanding, diagrams of components other than those necessary for the explanation have been omitted as appropriate.

[0019] (1) Overall configuration of the motion recognition system Figure 1 is a block diagram showing the functional configuration of the motion recognition system 1 according to this embodiment, Figure 2 is a diagram showing an example of an operation pattern for a specific operation in this embodiment, Figure 3 is a block diagram showing the functional configuration of the 3D sensor 20, Figure 4A is a schematic diagram showing the dot pattern irradiation of the 3D sensor 20, and Figure 4B is a diagram showing an example of the measurement range of the 3D sensor 20. The overall configuration of the motion recognition system 1 will be described below with reference to the diagrams.

[0020] The motion recognition system 1 according to this embodiment is a system that selects a specific motion pattern of a target person and displays an alert when a specific motion occurs while the person is being imaged. As shown in Figure 1, the motion recognition system 1 includes a specific motion data storage unit 10 that stores specific motion data indicating a specific motion including a predetermined series of postures and movements of at least one body part of the target M; a 3D sensor 20 as a detection means that continuously acquires point cloud data P indicating the 3D coordinate position on the surface of the target M and detects the motion of the target M from the point cloud data P; and a display unit 30 as a display means that continuously displays the motion of the target M detected by the 3D sensor 20 as a moving image. The motion recognition system 1 also includes a motion estimation unit 40 as an estimation means that estimates the motion of the target M from the continuous changes in the point cloud data P; an input reception unit 50 as a reception means that accepts the specification of a specific motion from among a plurality of motions of the target M; a notification unit 60 as a notification means that notifies that a specific motion has occurred when the motion of the target M matches the specified specific motion; and a recording unit 70 that records the point cloud data P acquired by the 3D sensor 20.

[0021] In the motion recognition system 1, the processing of the specific motion data storage unit 10, display unit 30, motion estimation unit 40, input reception unit 50, notification unit 60, and recording unit 70 is implemented, for example, by a general computer in which a processor consisting of a CPU (Central Processing Unit), ROM (Read Only Memory), and RAM (Random Access Memory) executes a predetermined program. The 3D sensor 20 can perform light emission control, light reception control, and communication control using a program processing unit implemented as a single-board computer equipped with an ARM processor, such as a Raspberry Pi (registered trademark), or a small computer device, but is not particularly limited.

[0022] (Specific operation data storage unit) The specific motion data storage unit 10 stores specific motions, including a predetermined series of postures and movements of body parts of the target M, as motion pattern data. Examples of specific motions include, but are not limited to, lying in bed, sitting up halfway out of bed, sitting on the bed, standing up from bed, touching one's arm, touching one's neck, touching one's shoulder, and standing, as shown in Figure 2 as an example of a motion pattern.

[0023] (3D sensor) As shown in the block diagram of Figure 3, the 3D sensor 20 comprises a projector 21 consisting of a light-emitting element that emits light, an image sensor module 22 consisting of an image sensor that receives reflected light reflected from the surface of the object to be imaged, and a control unit 23 that controls the light emission of the projector 21, the light reception of the image sensor module 22, and the processing of the received light data.

[0024] The projector 21 consists of a VCSEL (Vertical Cavity Surface Emitting Laser) 211, which is an array light source composed of multiple light-emitting elements, and a transmissive diffractive optical element (DOE) 212 that controls the light distribution of the laser beam above the VCSEL 211. The VCSEL211 emits invisible near-infrared laser light with a wavelength of 940 nm. Therefore, imaging is possible regardless of whether it is daytime or nighttime. The diffractive optical element 212 converts the light emitted by the VCSEL 211 into a regular dot pattern of light distribution and illuminates the target M as multiple beam-shaped ranging beams.

[0025] Figure 4A shows an example in which the diffractive optical element 212 converts the light emitted from the projector 21 into ranging beams L1 to Ln with a spread angle θ so that they are distributed to the target M as a regular dot pattern. In this embodiment, the dot pattern is irradiated with 4000 to 5000 laser beams, and as an example, as shown in Figure 4B, the measurement distance (L) is 300 mm to 2000 mm and the measurement range (R) is 285 × 285 mm to 1900 × 1900 mm.

[0026] The image sensor module 22 uses a CMOS (Complementary Metal Oxide Semiconductor) image sensor, which is a photoelectric conversion element. The image sensor module 22 receives the reflected light from the target M when invisible near-infrared laser light emitted from the projector 21 is reflected, and acquires point cloud data P.

[0027] In this embodiment, the control unit 23 is composed of a Raspberry Pi (registered trademark), which is a single-board computer equipped with an ARM processor. The control unit 23 controls the operation of the projector 21 and the image sensor module 22, and processes the point cloud data P acquired by the image sensor module 22 to generate 3D coordinate information values.

[0028] The 3D sensor 20 configured in this way emits invisible laser light from the projector 21 onto the target M, and while receiving the reflected light reflected from the surface of the target M with the image sensor module 22, it acquires point cloud data P and detects the three-dimensional movement of the target M.

[0029] (Display) The display unit 30 can be implemented in various forms, but in this embodiment, it can be implemented as the display of a computer on which the program for the motion recognition system 1 is installed, and outputs various information. In this embodiment, the motion of the target M detected by the 3D sensor 20 is displayed in real time as point cloud data P having 3D coordinate information values. The display unit 30 may also be configured as a touch panel display and serve as the input reception unit 40.

[0030] The operation of the target M displayed on the display unit 30 is displayed as point cloud data P having three-dimensional coordinate information values. Therefore, compared to camera images (RGB color images) using a webcam capable of capturing both still and moving images, the risk of acquiring personally identifiable information is lower, resulting in a display that takes privacy protection into consideration.

[0031] (Motion estimation part) The motion estimation unit 40 estimates the motion of the target M from the continuous changes in the point cloud data P acquired by the 3D sensor 20. Specifically, it estimates the motion of the target M using a learning model that has been trained on point cloud data P annotated to indicate the motion of the target M as training data. It is preferable to perform annotation work in advance and create a learning database as training data. For example, continuous changes (video) in the point cloud data P include "touching the body with the hand," "difficult to determine whether or not it is touching," "the state of the body changes," and "another object is visible," and annotations are added to these intervals.

[0032] (Input reception section) The input receiving unit 50 receives the specification of a specific operation from among multiple operations of the target M. Specifically, for example, it receives input information of a specific operation pattern specified by the user of the operation recognition system 1. The input operation of the input information is performed through the input device of the computer that constitutes the operation recognition system 1. For example, if the computer that constitutes the operation recognition system 1 is equipped with a touch panel display, the touch panel display functions as the display unit 30 and the input receiving unit 50. Note that the input receiving unit 50 is not limited to a touch panel display, but may be, for example, a keyboard, mouse, or mechanical switch. The input receiving unit 50 may also be a microphone. Alternatively, the input receiving unit 50 may be part of an independent device, such as a remote controller, or it may be part of a mobile terminal (smartphone or tablet) with a dedicated application program installed. In that case, the input receiving unit 50 receives input information by receiving it from the mobile terminal via the network.

[0033] The specific action patterns specified by the user are those pre-stored in the specific action data storage unit 10. Examples include lying in bed, sitting up halfway out of bed, sitting on the bed, standing up from bed, touching one's arm, touching one's neck, touching one's shoulder, and standing. Furthermore, specific operation patterns pre-stored in the specific operation data storage unit 10 can be added as operations that require special attention according to the situation of each target M, and the user can specify the added specific operation patterns via the input reception unit 50.

[0034] (News Department) The notification unit 60 issues an alert indicating that a specified operation has occurred when the operation of the target M matches the operation specified by the unit. In this context, an action of target M that matches a specified action is an action of the detected target M that is judged to have a certain degree of probability of being an action designated by the user as a specific action that should be detected, prompting the user to confirm the situation. For example, the system determines whether a specific action may have occurred, such as sitting on the bed, standing up from the bed, touching one's arm, touching one's neck, touching one's shoulder, or standing up. If it is determined that a specific action has occurred, the notification unit 60 notifies that a specific action has occurred.

[0035] The notification unit 60 may consist of, for example, a speaker that notifies information by sound, a vibrator that notifies information by vibration, or a lamp that notifies information by lighting up or flashing. Alternatively, the display unit 30 may function as the notification unit 60 by notifying information by displaying it on a display. The notification unit 60 is not limited to the example shown herein and may consist of other mechanisms capable of broadcasting information, or may be composed of multiple combinations thereof. Furthermore, these notification units 60 may be provided independently of the computer constituting the motion recognition system 1 and controlled by communication.

[0036] When the notification unit 60 notifies that a specific operation has occurred, it records the point cloud data P acquired by the 3D sensor 20. The recording unit 70 that records the point cloud data P can be composed of, for example, a semiconductor memory including non-volatile memory such as NVRAM (Non-Volatile RAM) or a recording medium such as an HDD (Hard Disk Drive). The recorded point cloud data can be shared among users by recording it as data in formats such as CSV (Comma Separated Value), JSON (JavaScript Object Notation), and XML (Extensible Markup Language).

[0037] (2) Detection process for specific actions Figure 5 is a schematic diagram showing an example of the arrangement of the 3D sensors 20 provided in the motion recognition system 1, Figure 6 is a flowchart showing the flow of the detection process for detecting the three-dimensional movement of the target M in the motion recognition system 1, Figure 7 is a conceptual diagram showing an example of the process of extracting point cloud data P of only the patient from the acquired point cloud data P, and Figure 8 is a conceptual diagram explaining an example of the process of removing noise from point cloud data P. The detection process of specific movements in the motion recognition system 1 according to this embodiment will be described below with reference to the drawings.

[0038] In the motion recognition system 1 according to this embodiment, as shown in Figure 5, the 3D sensor 20 is installed above the bed 80 and is configured to capture images mainly of the state and movements of patient M, which is an example of a target M using the bed 80.

[0039] First, in step S101, the 3D sensor 20 is turned on, and laser light is shone from the projector 21 onto patient M (S101). The laser beam is emitted as thousands of laser beams aligned simultaneously as dots. The laser beam is emitted at regular intervals while acquiring the point cloud data P of patient M, and the laser beam emitted as aligned dots toward patient M is received by the image sensor module 22 and acquired as point cloud data P in real time (S102).

[0040] Next, point cloud data P representing the background other than bed 80 and patient M is removed from the point cloud data P acquired in step S102, and point cloud data P representing only bed 80 and patient M is extracted (S103). As an example, as shown in Figure 7A, several tens of points are extracted from the point cloud data P acquired for each frame, in the direction of the X and Y axes, and the variance σ1 in the Z-axis direction, which is orthogonal to the X and Y axes, is calculated. If the calculated variance σ1 in the Z-axis direction is greater than a predetermined value, it can be estimated that there is movement, i.e., that patient M has been detected, and the point cloud data P of patient M can be extracted (see Figure 7B), but this method is not limited to this. In addition, for bed 80, the area of ​​bed 80 is calculated from the size of bed 80 that has been registered in advance, and point cloud data indicating bed 80 and bed 80's position is extracted (S103).

[0041] In step S103, noise is removed from the point cloud data P that shows the background other than bed 80 and patient M, and the point cloud data P of bed 80 and patient M is extracted (S104). As schematically shown in Figure 8, the variance σ2 of neighboring points in the point cloud data P that fall outside a virtual sphere with a predetermined threshold radius centered on any point Pn in the acquired point cloud data P is calculated. If the variance σ2 is greater than a predetermined value, it is judged as noise and removed from the point cloud data P. This process is repeated for each point in the point cloud data P to remove noise from the point cloud data P and extract the point cloud data P of bed 80 and patient M (S104). Alternatively, noise may be detected by calculating statistical outliers where the average distance from any point Pn in the acquired point cloud data P to neighboring points is far from the average of the entire point cloud data, and this noise may be removed from the acquired point cloud data P. This process may also be performed additionally. Note that if the point cloud data P of bed 80 and patient M can be distinguished from noise with a certain degree of confidence, the noise reduction process in step S104 may be skipped.

[0042] Then, in step S105, the three-dimensional posture of patient M is estimated from the point cloud data P of the region where bed 80 and patient M are located (S105). Here, Figure 9 shows an example of calculating the 3D coordinate position on the surface of patient M from point cloud data P using trigonometry. In Figure 9, the distance D between any point Pn in the point cloud data P and the 3D sensor 20 is given by Equation 1, where A is the distance between the projector 21 and the image sensor module 22, α1 is the angle formed between the measuring light L1 and the central axis c1 of the projector 21, and α2 is the angle formed between the central axis c2 of the image sensor module 22 and the reflected light R1 received by the image sensor module 22. D*tanα1 + D*tanα2 = A (Equation 1) This allows for the calculation of the distance from the projector 21 to the patient M onto which the ranging light L1 is projected, i.e., the three-dimensional coordinate position on the surface of the patient M. For each point in the point cloud data P of bed 70 and patient M, the three-dimensional coordinate position can be calculated from the above relational expression.

[0043] The 3D coordinate positions of the point cloud data P calculated in this way represent the state of patient M relative to bed 80, and the actions of patient M can be estimated from the continuous changes in the 3D coordinate positions of each point cloud data P. For example, using a learning model trained on training data annotated with "touching the body with hands" intervals as a continuous change (video) of point cloud data P, it can be estimated that actions such as touching the arm, touching the neck, and touching the shoulder occurred in patient M.

[0044] Next, the specific motion data storage unit 10 determines whether the patient M's state and actions, as estimated by the motion estimation unit 40, are specific actions designated by the user, based on the specific motion data stored in the storage unit 10 (step S106). Specifically, the system determines whether patient M is in one of the following states: lying in bed 80 (supine or lateral position), sitting on the edge of the bed (sitting on the edge), touching their arm, getting up from bed (getting out of bed), touching their arm, touching their neck, touching their shoulder, or standing. Whether a particular action is occurring is determined by whether the difference between the stored specific action data and the 3D coordinate position of the action estimated by the action estimation unit 40 is smaller than a predetermined threshold.

[0045] In step S106, if it is determined that a specific action has occurred in patient M during imaging (step S106; Yes), the notification unit 60 issues an alert indicating that a specific action has occurred (step S107). Specifically, the patient M's condition and movements are displayed on the screen as point cloud data P in real time, and notifications are made by displaying on the screen that it is a specified movement, turning on or flashing lights, or emitting sound through a speaker (S107).

[0046] Then, when the notification unit 60 notifies that a specific operation has occurred, in step S108, the point cloud data P acquired by the 3D sensor 20 is recorded (step S108). The point cloud data P to be recorded is recorded as data in formats such as CSV (Comma Separated Value), JSON (JavaScript Object Notation), and XML (Extensible Markup Language). This allows specific actions performed by patient M to be shared among users utilizing the motion recognition system 1.

[0047] As described above, the motion recognition system 1 according to this embodiment includes: a specific motion data storage unit 10 that stores specific motion data indicating a specific motion including a predetermined series of postures and movements of at least one body part of a target person; a 3D sensor 20 that continuously acquires point cloud data P indicating the 3D coordinate position on the surface of a target person and detects the person's movements from the point cloud data P; a display unit 30 that continuously displays the person's movements detected by the 3D sensor 20 as point cloud data P; a motion estimation unit 40 that estimates the target person's movements from the continuous changes in the point cloud data P; an input reception unit 50 that accepts the designation of a specific motion from among a plurality of movements of a target person; a notification unit 60 that notifies that a specific motion has occurred when the target person's movement matches the designated specific motion; and a recording unit 70.

[0048] With this configuration, the point cloud data P, which has a three-dimensional coordinate position, allows for real-time monitoring of various three-dimensional movements of the patient M. In particular, each point in the point cloud data P has a recorded three-dimensional coordinate position, making it possible to recognize the movements of body parts that are in close contact with the body and are difficult to perceive. Furthermore, with this configuration, the movements of patient M displayed on the display unit 30 are displayed as point cloud data P with three-dimensional coordinate positions. Compared to camera images (RGB color images) using a webcam, this reduces the risk of acquiring personally identifiable information, resulting in a display that takes privacy into consideration.

[0049] In this embodiment, an example of detecting the state and specific actions of patient M using bed 70 has been described. However, the motion recognition system 1 according to this embodiment can be suitably installed in hospitals, nursing homes, elderly care facilities, and residential homes, etc., depending on the type of object for which privacy protection is required, and can recognize and analyze the state and actions of a person who needs to be monitored, for example. [Explanation of symbols]

[0050] 1. Motion recognition system 10. Specific operation data storage unit 20.3D sensors 21...Projector, 211...VCSEL, 212...Diffractive optical element 22. Image sensor module 23.. Processing Unit 30...Display section 50...Input reception section 60... News Department 70... Records Department 80...bed M...Target (Patient)

Claims

1. A motion recognition system that recognizes the actions of subjects for whom privacy protection is required, A storage means for storing specific action data that indicates a specific action including a predetermined series of postures and movements of at least one body part of the aforementioned object, The system comprises an array light source consisting of multiple light-emitting elements that project invisible light as multiple beam-shaped ranging lights by converting it into a regular dot pattern light distribution pattern with a divergence angle via a diffractive optical element, and a light-receiving means that receives reflected light reflected from the surface of the object. A detection means that does not acquire RGB images, but acquires only point cloud data indicating the three-dimensional coordinate position on the surface of the object in real time, and detects the three-dimensional movement of the object from the point cloud data, A display means that displays the movement of the target detected by the detection means as point cloud data in real time as a moving image, Estimation means for estimating the movement of the object from the continuous changes in the point cloud data, A receiving means for receiving the designation of a specific operation among the multiple operations of the target, A notification means that notifies that the specified operation has occurred when the operation of the target is an operation that matches the specified operation, The system includes a recording means for sequentially saving the point cloud data acquired by the detection means, A motion recognition system characterized by the following features.

2. The estimation means uses a learning model trained on annotated training data to estimate the behavior of the target. The motion recognition system according to feature 1.

3. The aforementioned specific action is a body movement that includes the angle and width of movement of the body part of the target. The motion recognition system according to claim 1 or 2, characterized in that it is the same as described in claim 1 or 2.

4. A motion recognition method for recognizing the actions of subjects whose privacy protection is required, A storage step for storing specific action data that indicates a specific action including a predetermined series of postures and movements of at least one body part of the subject, The process involves converting an array light source consisting of multiple light-emitting elements into a regular dot pattern light distribution pattern with a divergence angle via a diffractive optical element, and projecting invisible light as multiple beam-shaped ranging lights. A light receiving step of receiving reflected light reflected from the surface of the object, A detection step in which, without acquiring an RGB image, only point cloud data indicating the three-dimensional coordinate position on the surface of the object is acquired in real time, and the three-dimensional movement of the object is detected from the point cloud data, A display step which displays the movement of the target detected in the detection step as point cloud data in real time as a moving image, An estimation step of estimating the movement of the object from the continuous changes in the point cloud data, A reception step that accepts the designation of a specific operation from among the multiple operations of the target, A notification step in which, if the operation of the target is an operation that matches the specified specific operation, the notification is made that the specific operation has occurred. A recording step includes sequentially saving the point cloud data acquired in the detection step, A method for recognizing motion characterized by the following features.

5. On the computer, A storage step of storing specific action data that indicates specific actions including a predetermined series of postures and movements of at least one body part of an object whose privacy is required, The process involves converting an array light source consisting of multiple light-emitting elements into a regular dot pattern light distribution pattern with a divergence angle via a diffractive optical element, and projecting invisible light as multiple beam-shaped ranging lights. The steps include receiving reflected light reflected from the surface of the object, A detection step in which, without acquiring an RGB image, only point cloud data indicating the three-dimensional coordinate position on the surface of the object is acquired in real time, and the three-dimensional movement of the object is detected from the point cloud data, A display step which displays the movement of the target detected in the detection step as point cloud data in real time as a moving image, An estimation step of estimating the specific action of the target from the continuous changes in the point cloud data, A reception step that accepts the designation of a specific operation from among the multiple operations of the target, A notification step in which, if the operation of the target is an operation that matches the specified specific operation, the notification is made that the specific operation has occurred. A recording step is performed to sequentially save the point cloud data acquired in the detection step. A program characterized by the following features.

Images