KR102825582B1 - Method and system for early prediction of self-harm and harm to others accidents based on artificial intelligence - Google Patents

Abstract

The present invention relates to an artificial intelligence-based method and system for early prediction of self-harm and harm to others. More specifically, the artificial intelligence-based method for early prediction of self-harm and harm to others according to the present invention may include a step of receiving sensing data on a dangerous patient sensed from at least one sensor, a step of analyzing the sensing data using an artificial intelligence model learned to predict signs of dangerous behavior related to self-harm or harm to others, and a step of predicting signs of the dangerous behavior related to either self-harm or harm to others of the dangerous patient using a result of the analysis through the artificial intelligence model.

Description

Method and System for Early Prediction of Self-Harm and Harm to Others Accidents Based on Artificial Intelligence {METHOD AND SYSTEM FOR EARLY PREDICTION OF SELF-HARM AND HARM TO OTHERS ACCIDENTS BASED ON ARTIFICIAL INTELLIGENCE}

The present invention relates to an artificial intelligence-based method and system for early prediction of self-harm and harm to others.

Medical institutions are implementing various preventive activities and countermeasures to ensure the safety of patients and medical staff. For example, they are implementing preventive activities such as eliminating risk factors in the surrounding environment and facilities to prevent suicide, self-harm, homicide, and other-injury accidents, and periodically assessing patients' depression and anxiety levels.

However, despite these efforts, suicide, self-harm, homicide, and harm to others continue to occur in medical institutions.

In particular, according to statistics from the Patient Safety Reporting Learning System, suicide, self-harm, homicide, and harm to others are steadily increasing, and incidents of patients or guardians committing violence against medical staff are also occurring frequently, which is emerging as a serious problem that threatens the safety and lives of medical staff.

Meanwhile, current response methods to self-harm and harm to others mainly rely on emergency bells and the deployment of security personnel.

However, these response methods are effective only when an emergency situation has already occurred or is imminent, so they have limitations in terms of preventing accidents. In addition, most of the existing studies targeting patients with depression or repeated self-harm have relied on patients' self-reports, which is less effective because it requires the cooperation of patients.

That is, because impulsive behaviors related to self-harm or harm to others can change very quickly, the need for a system that can detect them in real time is emerging.

Therefore, in order to ensure the safety of patients and medical staff within medical institutions, there is a need for a system that can predict the risk of self-harm and harm to others in advance in all medical fields, including psychiatry.

The present invention provides an artificial intelligence-based method and system for early prediction of self-harm and harm to others, which can monitor in real time the risks associated with self-harm or harm to others and prevent them.

More specifically, the present invention is to provide an artificial intelligence-based method and system for early prediction of self-harm and harm to others, which can predict risk behaviors related to self-harm or harm to others in a patient at an early stage and proactively respond to the predicted risk behaviors.

Furthermore, the present invention provides an artificial intelligence-based method and system for early prediction of self-harm and harm to others, which enables early prediction of risky behaviors related to self-harm or harm to others and immediate preventive measures to be taken.

Furthermore, the present invention provides an artificial intelligence-based method and system for early prediction of self-harm and harm to others, which can generate and provide response measures according to various risk factors so as to efficiently manage the resources of medical institutions.

In order to solve the problem discussed above, the method for early prediction of self-harm and harm to others based on artificial intelligence according to the present invention may include the steps of receiving sensing data on a risk patient sensed from at least one sensor, analyzing the sensing data using an artificial intelligence model learned to predict signs of risky behavior related to self-harm or harm to others, and predicting signs of the risky behavior related to either self-harm or harm to others of the risk patient using the analysis result through the artificial intelligence model.

Furthermore, the sensing data may include biometric information of the at-risk patient and an image captured of the at-risk patient, and the biometric information may include at least one of body temperature, pulse, blood pressure, respiratory rate, and skin tension of the at-risk patient.

Furthermore, in the step of analyzing the sensing data, the behavior of the at-risk patient can be analyzed from the image using the artificial intelligence model, and the behavior information of the at-risk patient can be extracted using the analysis result of the behavior of the at-risk patient.

Furthermore, in the step of analyzing the sensing data, the gaze and viewing pattern of the at-risk patient can be analyzed from the image using the artificial intelligence model, and the gaze tracking information of the at-risk patient can be extracted using the analysis result of the gaze and viewing pattern of the at-risk patient.

Furthermore, the gaze tracking information may include at least one of the gaze movement path, gaze target, gaze time, and gaze frequency of the at-risk patient.

Furthermore, the signs of the risky behavior may further include a step of predicting the risk level according to the signs of the risky behavior among a plurality of risk levels by using the biometric information, the behavioral information, and the eye tracking information of the risk patient, and a step of generating countermeasure information to prevent the self-harm or the harm to others accident related to the signs of the risky behavior based on the specified risk level.

Furthermore, the plurality of risk levels include a first risk level and a second risk level that are specified from signs of different risk behaviors, wherein the signs of different risk behaviors include signs of risk behavior related to the self-harm incident and signs of risk behavior related to the other-harm incident, and the first risk level can be specified based on the signs of risk behavior related to the other-harm incident being predicted through the artificial intelligence model, and the second risk level can be specified based on the signs of risk behavior related to the self-harm incident being predicted through the artificial intelligence model.

Furthermore, for each of the plurality of risk levels, different countermeasure information is matched and present, and the different countermeasure information includes first countermeasure information matched to the first risk level and second countermeasure information matched to the second risk level, and in the step of generating the countermeasure information, if the specific risk level corresponds to the first risk level, the first countermeasure information for preventing the third-party harm accident can be generated, and if the specific risk level corresponds to the second risk level different from the first risk level, the second countermeasure information for preventing the self-harm accident can be generated.

Meanwhile, the artificial intelligence-based early prediction system for self-harm and harm to others according to the present invention includes a communication unit that receives sensing data on a dangerous patient sensed from at least one sensor, and a control unit that analyzes the sensing data using an artificial intelligence model learned to predict signs of dangerous behavior related to self-harm or harm to others, wherein the control unit can predict signs of the dangerous behavior related to either self-harm or harm to others of the dangerous patient using the analysis result through the artificial intelligence model.

Meanwhile, a program according to the present invention is a program that is executed by one or more processes in an electronic device and can be stored in a computer-readable recording medium, wherein the program may include commands for performing a step of receiving sensing data on a dangerous patient sensed from at least one sensor, a step of analyzing the sensing data using an artificial intelligence model learned to predict signs of dangerous behavior related to self-harm or harm to others, and a step of predicting signs of the dangerous behavior related to either self-harm or harm to others of the dangerous patient using a result of the analysis through the artificial intelligence model.

As examined above, the artificial intelligence-based self-harm and other-harm accident early prediction method and system according to the present invention can contribute to preventing unnecessary accidents and efficiently managing the resources (e.g., manpower, time, cost, etc.) of medical institutions by early predicting risky behaviors related to self-harm or other-harm accidents.

In addition, the method and system for early prediction of self-harm and harm to others based on artificial intelligence according to the present invention can contribute to early prediction and prevention of risk behaviors related to self-harm or harm to others of at-risk patients and to establishing a medical service environment that is safe for patients and medical staff and helpful in increasing treatment effectiveness by building a data integration management platform with information sensed from various sensors and immediately applicable to the field.

Furthermore, the artificial intelligence-based method and system for early prediction of self-harm and harm to others according to the present invention can provide a medical service environment in which medical staff working at medical institutions can protect themselves from patients at high risk of harm to others by conducting a linkage analysis of the mutual relationship between at-risk patients and specific targets.

Furthermore, the artificial intelligence-based method and system for early prediction of self-harm and harm to others according to the present invention can enable the development of various digital therapeutics targeting self-harm and harm to others by constructing a database based on sensing data related to self-harm and harm to others.

Furthermore, the AI-based self-harm and other-harm accident early prediction method and system according to the present invention can early predict a patient's risk behavior related to self-harm or other-harm accidents and generate and provide response measure information that can proactively respond to the predicted risk behavior. Through this, medical institutions can drastically reduce the economic burden by reducing the demand for personnel such as safety personnel in the long term.

FIGS. 1 and 2, 3a and 3b are conceptual diagrams for explaining an artificial intelligence-based method and system for early prediction of self-harm and harm to others according to the present invention.
Figure 4 is a flowchart for explaining an artificial intelligence-based method for early prediction of self-harm and harm to others according to the present invention.
FIG. 5a, FIG. 5b, FIG. 6, FIG. 7, FIG. 8, and FIG. 9 are conceptual diagrams for explaining an artificial intelligence-based method for early prediction of self-harm and harm to others according to the present invention.

Hereinafter, embodiments disclosed in this specification will be described in detail with reference to the attached drawings. Regardless of the drawing symbols, identical or similar components will be given the same reference numerals and redundant descriptions thereof will be omitted. The suffixes "module" and "part" used for components in the following description are assigned or used interchangeably only for the convenience of writing the specification, and do not have distinct meanings or roles in themselves. In addition, when describing embodiments disclosed in this specification, if it is determined that a specific description of a related known technology may obscure the gist of the embodiments disclosed in this specification, the detailed description thereof will be omitted. In addition, the attached drawings are only intended to facilitate easy understanding of the embodiments disclosed in this specification, and the technical ideas disclosed in this specification are not limited by the attached drawings, and should be understood to include all modifications, equivalents, and substitutes included in the spirit and technical scope of the present invention.

Terms including ordinal numbers, such as first, second, etc., may be used to describe various components, but the components are not limited by the terms. The terms are used only to distinguish one component from another.

When it is said that a component is "connected" or "connected" to another component, it should be understood that it may be directly connected or connected to that other component, but that there may be other components in between. On the other hand, when it is said that a component is "directly connected" or "directly connected" to another component, it should be understood that there are no other components in between.

Singular expressions include plural expressions unless the context clearly indicates otherwise.

In this application, it should be understood that terms such as “comprises” or “have” are intended to specify the presence of a feature, number, step, operation, component, part or combination thereof described in the specification, but do not exclude in advance the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts or combinations thereof.

The present invention relates to an artificial intelligence-based method and system for early prediction of self-harm and harm to others, which can predict (or in advance) risk behavior related to self-harm or harm to others of at-risk patients by analyzing information collected from various sensors.

Self-harm can refer to any behavior that an individual does to harm their own body or mind. It can be an intentional act of damaging one's own body tissues and can be used as a way to relieve emotional pain or emotional release. Examples of self-harm include physical harm (e.g., cutting or scratching oneself with a knife or sharp object, banging one's head against a wall, etc.), substance abuse (e.g., deliberately consuming excessive amounts of drugs or alcohol), and self-poisoning (e.g., ingesting toxic substances to harm one's body).

Harm to Others can refer to any behavior that causes physical, mental, or physical harm to another person. Examples of harm to others include physical violence (e.g., hitting or physically harming another person), psychological violence (e.g., intentionally causing psychological distress to another person), and material harm (e.g., intentionally damaging or destroying another person's property).

Meanwhile, as illustrated in FIG. 1, the “various sensors” described in the present invention may include a contact sensor (1) and a non-contact sensor (2).

The contact sensor (1) may include sensors that can sense (or measure) the patient's bio-information (e.g., body temperature, pulse, blood pressure, respiration rate, skin tension, etc.) by coming into contact with the patient's body. In this case, the contact sensor may include a wearable device consisting of a smart watch. For example, the contact sensor may include a temperature sensor, a heart rate sensor, a blood pressure sensor, a respiration rate sensor, a skin tension sensor, an electrocardiogram sensor, an SPO2 sensor, a galvanic skin response sensor, an accelerometer, a gyroscope, an electromyography sensor, etc.

The non-contact sensor (2) may include sensors that can sense various information related to a patient (e.g., vital signs (or biometric information), location information, image data, 4D radar, etc.) without coming into contact with the patient's body. For example, it may include a camera (e.g., RGB camera, IR camera, etc.), CCTV, imaging sensor, infrared sensor, radar sensor, LiDAR sensor (Light Detection and Ranging sensor), ultrasonic sensor, 4D radar sensor, microwave sensor, optical sensor, laser Doppler sensor, position sensor, robot, etc.

However, the sensors included in the contact sensor (1) and the non-contact sensor (2) are not necessarily limited to this, and may include various types of sensors in addition to the sensors mentioned above.

Furthermore, the “at-risk patient” described in the present invention may mean a patient who feels pain due to a mental illness, and in the present invention, the “at-risk patient” may be used interchangeably with “patient” or “user.”

In the present invention, for the convenience of explanation, the description focuses on “mental illness”, but is not necessarily limited thereto. That is, the at-risk patients described in the present invention may include not only patients with mental illness (e.g., mentally ill patients), but also patients with various diseases (e.g., chronic pain, alcoholism, drug abuse and dependence, etc.).

Referring again to FIG. 1, the artificial intelligence-based self-harm and other-harm accident early prediction method and system (hereinafter, “self-harm and other-harm accident early prediction system, 100) according to the present invention can construct a smartwatch-type device capable of comprehensively applying various contact sensors (1), and a sensor-based real-time surveillance system capable of collecting image data and location information data to which anonymization for personal information protection has been applied.

As an example, the early prediction system for self-harm and harm to others (100) can build a sensor-based real-time surveillance system that enables patient recognition even at night, and can build a sensor-based real-time surveillance system that can detect dangerous behavior related to self-harm or harm to others attempted in a space where video acquisition is impossible, such as a bathroom or shower room (for example, a radar sensor can be attached to the ceiling to measure the distance to the head and detect danger when it gets close to a certain level (ex: 2 m or more from the ground).

Meanwhile, conventionally, data extracted from individual behaviors are used to predict risky behaviors, but most risky behaviors related to self-harm or harm to others are often caused by interactions between individuals or between individuals and groups rather than individual behaviors. Therefore, in order to predict self-harm or harm to others caused by interactions, it is necessary to collect and analyze image data of groups existing in close space as well as individual behaviors.

Accordingly, the self-harm and harm to others early prediction system (100) can implement (or develop) a more complete self-harm and harm to others early prediction algorithm through the collection and analysis of multiple multi-modal sensing (or sensor) data targeting interactions between individuals.

For example, camera-based behavior detection requires time series analysis time to secure reliability, and too fast alarms can reduce the reliability of the alarms due to frequent false judgments. Accordingly, the self-harm and other-harm accident early prediction system (100) can enhance reliability by deriving a more robust recognition algorithm that uses image analysis and multimodal data sets utilizing additional sensors. That is, the self-harm and other-harm accident early prediction system (100) can recognize (or predict) risky behaviors related to self-harm or other-harm accidents early by applying a robust recognition algorithm, and implement a risky behavior recognition algorithm that maintains reliability.

As another example, the early prediction system for self-harm and harm to others (100) can analyze signs of risky behaviors related to self-harm or harm to others using a multimodal algorithm. More specifically, the early prediction system for self-harm and harm to others (100) can secure a deep learning network model that has learned behavior recognition and biometric information data sets, and apply the secured network model to real-time field data to predict signs of risky behaviors.

As another example, the early prediction system for self-harm and harm to others (100) can be utilized to predict abnormal signs of at-risk patients (e.g., predict signs of at-risk behavior) by using an eye-tracking algorithm that analyzes gaze and gaze patterns of at-risk patients before their risky behavior. More specific details about the eye-tracking algorithm will be described later.

On the other hand, predicting risks related to self-harm or harm to others alone cannot be expected to have a significant preventive effect. In order to prevent self-harm or harm to others, it is necessary to establish an additional system that can proactively respond based on increased risk detected in the field.

Accordingly, the early prediction system for self-harm and harm to others (100) can attempt to simultaneously achieve rapid response and management efficiency by establishing a crisis management-based monitoring and rapid response system (e.g., real-time group alarm (or integrated alarm), mobile cognitive behavioral therapy (or responsive cognitive behavioral therapy) that can be proactively provided to patients with increased risk, and patrol using a circulating robot, etc.).

Furthermore, the early prediction system for self-harm and harm to others (100) can be implemented in various platform forms such as applications, software, and websites.

Below, the configuration of the self-harm and other-harm accident early prediction system (100) discussed above will be described in more detail with the attached drawings. Figure 2 is a conceptual diagram for explaining the artificial intelligence-based self-harm and other-harm accident early prediction method and system according to the present invention.

As illustrated in FIG. 2, the self-harm and harm to others accident early prediction system (100) according to the present invention may include at least one of a sensing unit (110), a communication unit (120), a storage unit (130), and a control unit (140).

The sensing unit (110) may include at least one sensor. More specifically, the sensing unit (110) is for sensing information related to a risk patient (U1) (or information about a space where a risk patient (U1) is located), and may include a contact sensor (111) and a non-contact sensor (112).

For example, the contact sensor (111) may include at least one of a temperature sensor, a heart rate sensor, a blood pressure sensor, a respiration rate sensor, a skin tension sensor, an electrocardiogram sensor, an oxygen saturation sensor, a galvanic skin response sensor, an accelerometer, a gyroscope, and an electromyography sensor.

For another example, the non-contact sensor (112) may include at least one of a camera, a CCTV, an imaging sensor, an infrared sensor, a radar sensor, a LiDAR sensor (Light Detection and Ranging sensor), an ultrasonic sensor, a 4D radar sensor, a microwave sensor, an optical sensor, a laser Doppler sensor, a position sensor, and a robot.

However, in the present invention, there is no particular limitation on the type of sensor constituting the sensing unit (110), and it is sufficient as long as the function defined by each sensor is implemented.

Next, the communication unit (120) may be connected to a control system (or control server, or control room, 1000), a device (or device), an external server, and one or more networks via a wireless or wired network to receive or transmit overall data and information necessary for the operation of the self-harm and other-harm accident early prediction system (100).

Here, the control system (1000) is configured to control patients located in a specific hospital (e.g., a mental hospital), and can manage patients using various information (e.g., information on response measures for at-risk patients (U1)) provided by the self-harm and harm to others early prediction system (100). However, in the present invention, for the convenience of explanation, the explanation is centered on a “hospital,” but is not necessarily limited thereto (e.g., it can be utilized in various places such as public transportation, companies, and public institutions).

In addition, the communication unit (120) can receive sensing data sensed from the sensing unit (110). For example, the communication unit (120) can receive biometric information, location information, and images of the at-risk patient (U1) sensed from the sensing unit (110).

Furthermore, the communication unit (120) can support various communication methods depending on the communication standards of the communicating device.

For example, the communication unit (120) may be configured to communicate with a communication target using at least one of the following technologies: WLAN (Wireless LAN), Wi-Fi (Wireless-Fidelity), Wi-Fi (Wireless Fidelity) Direct, DLNA (Digital Living Network Alliance), WiBro (Wireless Broadband), WiMAX (World Interoperability for Microwave Access), HSDPA (High Speed Downlink Packet Access), HSUPA (High Speed Uplink Packet Access), LTE (Long Term Evolution), LTE-A (Long Term Evolution-Advanced), 5G (5th Generation Mobile Telecommunication), Bluetooth™ (Bluetooth™), RFID (Radio Frequency Identification), Infrared Data Association (IrDA), UWB (Ultra-Wideband), ZigBee, NFC (Near Field Communication), Wi-Fi Direct, and Wireless USB (Wireless Universal Serial Bus).

Meanwhile, the storage unit (130) may also be called a “database (DB)” or a “memory” and may be configured to store various information related to the present invention. In the present invention, the storage unit (130) may be provided in the self-harm and other-harm accident early prediction system (100) itself. In addition, at least a part of the storage unit (130) may be configured as a cloud server (or cloud storage). That is, the storage unit (130) may be sufficient as a space where information (or data) necessary for the operation of the self-harm and other-harm accident early prediction system (100) according to the present invention is stored, and it may be understood that there are no restrictions on the physical space.

The storage unit (130) may store various information related to the at-risk patient (U1). For example, various information related to the at-risk patient (U1) may include the patient's face, voice, name, gender, age, disposition (or personality), disease (or illness), disease occurrence date, treatment (or care) history, occupation, interests, hobbies, etc.

Meanwhile, in the storage unit (130), information on at least one of biometric information, behavioral information, and eye tracking information related to self-harm or harm to others of a dangerous patient (U1) can be matched and stored.

First, in the case of a third-party accident, in addition to the dangerous patient (U1), there may be at least one of the third-party victims (e.g., medical staff, other patients, etc.) and specific objects (e.g., equipment, instruments, objects, things, etc.) that are the targets of the third-party accident. In this regard, the control unit (140) may analyze the relationship and vital signs between the dangerous patient (U1) and the third-party victims to extract (or collect) at least one of the vital information, behavioral information, and eye tracking information of the dangerous patient (U1) and the third-party victims.

For example, as illustrated in FIG. 3a, the control unit (140) can extract biometric information (ex: “body temperature: 38°C”, “pulse: 100 bpm”, “blood pressure: 140/90 mmHg”, “respiratory rate: 22 breaths per minute”, 311) of a dangerous patient (U1) and biometric information (ex: “body temperature: 37.5°C”, “pulse: 90 bpm”, “blood pressure: 120/80 mmHg”, “respiratory rate: 19 breaths per minute”, 321) of a target of harm (U2) before a dangerous behavior related to a harm accident occurs.

In addition, the control unit (140) can analyze the video (310) related to the accident involving a dangerous patient (U1) based on an artificial intelligence model (or artificial intelligence algorithm, 150) to collect behavioral information and eye tracking information before the occurrence of the dangerous behavior related to the accident involving a dangerous patient (U1). For example, the control unit (140) can extract, based on the analysis results, behavioral information (ex: “Continuously raises the right hand”, “Shakes both feet and both hands”, 312) before a dangerous behavior related to a harm accident of a dangerous patient (U1) occurs, eye tracking information (ex: “Gaze time: 5 seconds or more”, “Gaze target: Person wearing green clothes”, “Gaze movement path: Fixed on the face of the gaze target”, “Gaze frequency: Focuses only on the gaze target, does not gaze elsewhere”, 313) and behavioral information of a harm target (U2) (ex: “Makes a gesture with both hands to a person wearing black clothes”, 322), eye tracking information (ex: “Gaze time: less than 2 seconds”, “Gaze target: Person wearing black clothes”, “Gaze movement path: Looks at the gaze target once and then looks at the surrounding environment”, “Gaze frequency: High gaze frequency of the surrounding environment”, 323).

Next, in the case of a self-harm incident, in addition to the at-risk patient (U1), a specific object (e.g., a person, equipment, apparatus, object, thing, etc.) may exist. In this regard, if a specific object exists, the control unit (140) may analyze the relationship between the at-risk patient (U1) and the specific object to extract at least one of the biometric information, behavioral information, and eye tracking information of the at-risk patient (U1).

For example, as illustrated in FIG. 3b, the control unit (140) can extract biometric information (ex: “body temperature: 38°C”, “pulse: 100 bpm”, “blood pressure: 140/90 mmHg”, “respiratory rate: 24 breaths per minute”, 331) of a dangerous patient (U1) sensed and stored (or recorded) by the sensing unit (110) before a dangerous behavior related to self-harm occurs.

In addition, the control unit (140) can analyze the video (330) related to the self-harm incident of the at-risk patient (U1) based on the artificial intelligence model, and collect behavioral information and eye tracking information prior to the occurrence of the dangerous behavior related to the self-harm incident of the at-risk patient (U1). For example, the control unit (140) can extract behavioral information (ex: “sitting still and shaking in a silent state”, “looking at the target of gaze and then turning the head and continuously searching the surroundings”, 332) and eye tracking information (ex: “gaze time: 3 seconds or more”, “gaze target: white line”, “gaze movement path: looking at the target of gaze once and then looking at the surroundings”, “gaze frequency: repeating at 3 to 5 second intervals”, 333) prior to the occurrence of the dangerous behavior related to the self-harm incident of the at-risk patient (U1) based on the analysis result.

However, the biometric information, behavioral information, and eye tracking information related to self-harm or harm to others of the at-risk patient (U1) are not necessarily limited thereto. For example, the storage unit (130) may further include at least one of i) normal (or average) biometric information, behavioral information, and eye tracking information of the at-risk patient (U1), and ii) biometric information, behavioral information, and eye tracking information immediately after the occurrence of a risky behavior related to self-harm or harm to others of the at-risk patient (U1).

Furthermore, in the storage unit (130), in addition to the at-risk patient (U1) registered in the self-harm and harm-to-other early prediction system (100), information on at least one of biometric information, behavioral information, and eye tracking information related to self-harm or harm-to-other of other patients (or at-risk patients or users) can be matched and stored.

In this case, information related to at least one of the biometric information, behavioral information, and eye tracking information of other patients may correspond to at least one of, for example, i) information collected from a server (e.g., a hospital (or medical institution) server) linked to the self-harm and harm-to-other accident early prediction system (100), ii) information collected (or specified) by analyzing a video related to self-harm or harm-to-other accident collected from an external server (or website), iii) information collected from an external organization (e.g., a medical institution, a research institute, etc.), or iv) information collected from an accident record related to self-harm and harm-to-other accident.

As an example, in the present invention, if a new at-risk patient other than a at-risk patient (U1) registered in the early prediction system for self-harm and harm to others (100) exists in an image sensed from a sensing unit (110), it is possible to predict signs of dangerous behavior related to self-harm or harm to others of the new at-risk patient by using information related to at least one of the biometric information, behavioral information, and eye tracking information of the at-risk patient and other patients.

That is, in the present invention, a database can be constructed using biometric information, behavioral information, and eye tracking information related to self-harm or harm to others of a patient at risk (U1) and biometric information, behavioral information, and eye tracking information related to self-harm or harm to others of other patients.

Furthermore, data and commands required for the operation of the self-harm and harm to others accident early prediction system (100) according to the present invention may be stored in the storage unit (130). For example, a learning data set required for learning an artificial intelligence model (150) may be stored in the storage unit (130), and data, software, firmware, program code (Source Code), and commands processed or scheduled to be processed by the control unit (140) may be further stored.

Meanwhile, the control unit (150) may also be called a “processor” and may play a role in controlling the overall operation of the self-harm and other-harm accident early prediction system (100) related to the present invention. The control unit (140) may process signals, data, information, etc. input or output through the components discussed above, or may perform a series of data processing to provide or process appropriate information and functions to the user.

The control unit (140) can train the artificial intelligence model (150) using learning data (or a learning data set) to predict signs of risky behavior related to self-harm or harm to others of a dangerous patient (U1).

The learning data of the artificial intelligence model (150) may be diverse. As described above, the control unit (140) may train the artificial intelligence model (150) using the biometric information, behavioral information, and eye tracking information related to self-harm or harm to others of a dangerous patient (U1) stored in the storage unit (130) as learning data. As another example, the control unit (140) may train the artificial intelligence model (150) using the biometric information, behavioral information, and eye tracking information related to self-harm or harm to others of other patients stored in the storage unit (130) as learning data.

Hereinafter, based on the configuration of the AI-based self-harm and other-harm accident early prediction system discussed above, the AI-based self-harm and other-harm accident early prediction method according to the present invention will be described in more detail. FIG. 4 is a flowchart for explaining the AI-based self-harm and other-harm accident early prediction method according to the present invention, and FIGS. 5a, 5b, 6, 7, 8, and 9 are conceptual diagrams for explaining the AI-based self-harm and other-harm accident early prediction method according to the present invention.

Meanwhile, in the present invention, a process of receiving sensing data on a risk patient sensed from at least one sensor may be performed (S410, see FIG. 4).

The control unit (140) can control the sensing unit (110) to receive sensing data about a risk patient. Here, the sensing data can include biometric information of the risk patient (U1) and an image captured of the risk patient. For example, as illustrated in FIG. 5A, the control unit (140) can control the sensing unit (110) to receive (or acquire or collect) an image (510) captured of the risk patient (U1) and biometric information (511) of the risk patient sensed by at least one sensor included in the sensing unit (110).

In this case, the biometric information (511) of the at-risk patient (U1) may be information sensed from a contact sensor (111) in contact with the body of the at-risk patient (U1). The biometric information (511) may include at least one of the body temperature, pulse, blood pressure, respiratory rate, and skin tension of the at-risk patient (U1). However, the biometric information is not necessarily limited thereto, and may further include various information (e.g., blood oxygen saturation, electrocardiogram, blood sugar level, weight, physical activity level, sleep pattern, etc.) in addition to the above.

In addition, the image (510) captured of the at-risk patient (U1) may be information sensed from a non-contact sensor (112). At this time, the information sensed from the non-contact sensor (112) may further include, in addition to the image (510), location information and biometric information (or biometric signs) of the at-risk patient (U1), 4D radar captured of the at-risk patient, etc. That is, in the present invention, it may be possible to sense biometric information of the at-risk patient (U1) not only through the contact sensor (111) but also through the non-contact sensor (112).

In this regard, in the present invention, in addition to the control unit (140) receiving sensing data through the control of the sensing unit (110), sensing data related to a patient at risk (U1) can be received using various methods.

As an example, the control unit (140) may receive sensing data on a risk patient (U1) from the sensing unit (110) based on criteria set in the self-harm and harm to others early prediction system (100). Here, the criteria set in advance may include at least one of: i) when at least one element of the biometric information of the risk patient (U1) is equal to or higher than a preset risk value (e.g., pulse 100 bpm), ii) when a certain number (e.g., 10) or more people exist in the space (or location) where the risk patient (U1) is sensed, iii) when the contact sensor (111) in contact with the body of the risk patient (U1) is not recognized.

However, for convenience of explanation, the present invention does not limit the method by which sensing data is received to any one.

That is, the control unit (140) receives sensing data using various methods, thereby enabling more stable and accurate real-time monitoring, and furthermore, it can flexibly respond to changes in the environment or changes in the condition of a patient at risk (U), thereby enabling rapid and effective risk management.

Furthermore, if the image (510) taken of the dangerous patient includes, in addition to the dangerous patient (U1), other specific subjects (e.g., patients, medical staff, etc.) located in a close space with the dangerous patient (U1), the control unit (140) can receive sensing data for the other subjects together. For example, the control unit (140) can control the sensing unit (110) to receive biometric information (ex: “body temperature: 37°C”, “pulse: 90 bpm”, “blood pressure: 120/80 mmHg”, “respiratory rate: 20 breaths per minute”, 521) for the other subject (U2).

In this case, a specific target in the present invention may also be referred to as a “target of another,” and for convenience of explanation, the two terms will be used interchangeably without distinguishing between them hereinafter.

Meanwhile, in the present invention, a process of analyzing sensing data can be performed using an artificial intelligence model learned to predict signs of risky behavior related to self-harm or harm to others (S420, see FIG. 4).

As discussed above, the control unit (140) can train an artificial intelligence model (150) using various data related to self-harm or harm to others stored in the storage unit (130) as learning data.

The control unit (140) can perform analysis on sensing data using a pre-learned artificial intelligence model (150). More specifically, as shown in FIG. 5a, the control unit (140) can analyze an image (510) captured of a dangerous patient (U1) using a pre-learned artificial intelligence model (150).

First, the control unit (140) can analyze the behavior of a patient at risk (U1) by analyzing an image (510) using an artificial intelligence model (150).

As an example, since the image data is composed of continuous frames, the control unit (140) can extract the image (510) in frame units. Then, the control unit (140) can detect the main joint positions (or key points) of the at-risk patient (U1) in each frame using the artificial intelligence model (150), and analyze the behavior of the at-risk patient (U1) based on the detected key points. The key points are coordinate information indicating the posture or gestures of the at-risk patient (U1), and the behavior of the at-risk patient (U1) can be analyzed based on these.

As another example, the control unit (140) can detect the position of the at-risk patient (U1) in each frame of the image (510) and track the movement of the at-risk patient (U1) between consecutive frames using the artificial intelligence model (150). Then, the control unit (140) can analyze the movement of the tracked at-risk patient (U1) and recognize a specific behavior (e.g., grasping the body, shaking, a specific posture, etc.) using the artificial intelligence model (150). Furthermore, the control unit (140) can extract a feature vector (e.g., type, duration, frequency, etc. of the behavior) from the recognized specific behavior and analyze the behavior pattern of the at-risk patient (U1) based on the extracted feature vector.

Next, the control unit (140) can extract behavioral information of the at-risk patient (U1) by using the behavioral analysis results for the at-risk patient (U1). For example, as illustrated in Fig. 5a, the control unit (140) can extract (or specify) behavioral information (ex: “taking a behavior of continuously raising the right hand, 512) of the at-risk patient (U1) based on the behavioral analysis results of the at-risk patient (U1).

In addition, the control unit (140) can analyze the behavior of a specific subject (U2) included in the image (510) to extract behavior information of the specific subject (U2). For example, the control unit (140) can analyze the behavior of a specific subject (U2) using an artificial intelligence model (150) and extract behavior information of the specific subject (U2) (ex: “Making a gesture with both hands toward a person wearing black clothes, 522) based on the analysis result.

Meanwhile, the control unit (140) can analyze the gaze and gaze pattern of a risk patient (U1) from the image (510) using an artificial intelligence model (or an artificial intelligence model learned based on a gaze tracking algorithm or a vision tracking algorithm, 150).

As an example, the control unit (140) can detect the eye position of a risk patient (U1) from an image (510) using an artificial intelligence model (150), and track the direction of the gaze of the risk patient (U1) based on the detected eye position. Then, the control unit (140) can determine the gaze and fixation pattern of the risk patient (U1) using the tracked direction of the gaze.

And, the control unit (140) can extract the gaze tracking information of the risk patient (U1) by using the analysis result of the gaze and fixation pattern of the risk patient. For example, as illustrated in FIG. 5b, the control unit (140) can extract the gaze tracking information of the risk patient (U1) (ex: “Gaze time: 5 seconds or more”, “Gaze target: person wearing gray clothes”, “Gaze movement path: fixed on the face of the gaze target”, “Gaze frequency: focused only on the gaze target”, 531) based on the determined gaze and fixation pattern of the risk patient (U1).

Here, the gaze tracking information may include at least one of the gaze movement path, gaze target, gaze time, and gaze frequency of the at-risk patient (U1). At this time, in the case of the gaze target, in addition to a specific subject (U2), a specific object (e.g., equipment, apparatus, object, thing, etc.) may be included in the gaze target. However, the information included in the gaze tracking information is not necessarily limited to this, and in addition to what is mentioned, various information (e.g., gaze direction, gaze fixation time, surrounding search frequency, number of changes in gaze target, gaze visual information, characteristics of gaze target, etc.) may be further included.

Furthermore, the control unit (140) can analyze the gaze and fixation pattern of a specific subject (U2) included in the image (510) using the artificial intelligence model (150), and extract the gaze tracking information of the specific subject (U2). For example, as illustrated in FIG. 5b, the control unit (140) can extract the gaze tracking information of the specific subject (U2) (ex: “Gaze time: less than 2 seconds”, “Gaze target: person wearing black clothes”, “Gaze movement path: looks at the gaze target and surrounding environment alternately”, “Gaze frequency: high gaze frequency of the surrounding environment”, 541) through the analysis result of the gaze and fixation pattern of the specific subject (U2).

Meanwhile, if the distance from the image (510) to the at-risk patient (U1) is greater than a certain distance, the control unit (140) can predict the direction of the gaze of the at-risk patient (U1) using an artificial intelligence algorithm and extract gaze tracking information of the at-risk patient based on the predicted direction of the gaze.

Specifically, the control unit (140) can predict the direction of the gaze of the patient at risk (U1) based on the head pose estimation algorithm when the distance from the sensing unit (or non-contact sensor, 110) to the patient at risk (U1) in the image (510) is greater than or equal to a preset distance (ex: 2M). However, the above-described preset distance value is only an example, and the preset distance value is not necessarily limited thereto, and it is obvious that it may be changed in various ways.

And, the control unit (140) can determine the gaze and viewing pattern of the at-risk patient (U1) using the predicted gaze direction, and extract the gaze tracking information of the at-risk patient (U1) based on this.

Furthermore, the control unit (140) can extract the gaze tracking information of the specific subject (U2) through the process described above even when the distance between the at-risk patient (U1) and the specific subject (U2) located in the vicinity is greater than the preset distance from the image (510). For example, the control unit (140) can predict the gaze direction of the specific subject (U2) using a head pose estimation algorithm, and extract the gaze tracking information of the specific subject (U2) based on the predicted gaze direction (ex: “gaze time: less than 2 seconds”, “gaze target: person wearing black clothes”, “gaze movement path: looks at the gaze target and the surrounding environment alternately”, “gaze frequency: high gaze frequency of the surrounding environment”, 541).

That is, in the present invention, not only the biometric information and behavioral information of the at-risk patient (U1) but also the gaze tracking information is extracted, so that the information on the person or object interacting with the at-risk patient (U1) can be usefully utilized for behavior analysis and risk factor identification of the at-risk patient (U1).

Meanwhile, in the present invention, a process of predicting signs of risk behavior related to either self-harm or harm to others of a risk patient can be performed using analysis results through an artificial intelligence model (S430, see FIG. 4).

The control unit (140) can predict signs of risk behavior related to either self-harm or harm to others by a risk patient based on the analysis results through the artificial intelligence model (150).

Here, signs of risk behavior may include warning signs that foretell actions or incidents that threaten the health and safety of oneself or others, such as harm or self-harm. These signs of risk behavior may be physical, emotional, behavioral, or verbal.

As discussed above, the control unit (140) can extract biometric information, behavioral information, and eye tracking information of a risk patient (U1) using an artificial intelligence model (150). Accordingly, signs of risky behavior can be predicted using the biometric information, behavioral information, and eye tracking information of a risk patient (U1).

More specifically, the control unit (140) can perform a linkage analysis of the relationship between the at-risk patient (U1) and the specific subject (U2) based on the extracted biometric information, behavioral information, and eye tracking information of the at-risk patient (U1) and the biometric information, behavioral information, and eye tracking information of the specific subject (U2).

And, the control unit (140) can predict signs of risk behavior related to either self-harm or harm to others of the at-risk patient (U1) based on the results of the mutual relationship linkage analysis.

As an example, the control unit (140) may analyze the relationship between biometric information, behavioral information, and eye tracking information between a risk patient (U1) and a specific subject (U2) and determine that the information is related to a third-party accident, and may predict signs of risky behavior related to the third-party accident of the risk patient (U1).

As another example, the control unit (140) can use an artificial intelligence model (150) learned with data related to a risk patient (U1) to identify that the biometric information, behavioral information, and eye tracking information of the risk patient (U1) are related to a risk-related accident, and predict signs of risky behavior related to the risk-related accident of the risk patient (U1).

That is, since in the present invention, as discussed above, a database is constructed using biometric information, behavioral information, and eye tracking information related to self-harm or harm to others of a patient at risk (U1), and biometric information, behavioral information, and eye tracking information related to self-harm or harm to others of other patients, it is possible to predict whether the biometric information, behavioral information, and eye tracking information of a patient at risk (U1) exhibit signs of risky behavior related to either self-harm or harm to others.

Meanwhile, in the early prediction system (100) for self-harm and harm to others according to the present invention, multiple risk levels may be matched (or set) according to each of the signs of different risk behaviors.

Here, the multiple risk levels may include a first risk level and a second risk level, which are specified from signs of different risk behaviors. In addition, the signs of different risk behaviors may include signs of risk behaviors related to self-harm thoughts and signs of risk behaviors related to other-harm thoughts.

In this regard, the first risk level can be specified based on the prediction of signs of risky behavior related to harm through the artificial intelligence model (150). In addition, the second risk level can be specified based on the prediction of signs of risky behavior related to self-harm through the artificial intelligence model (150).

Meanwhile, when the control unit (140) completes predicting the signs of risk behavior related to either self-harm or harm to others of the at-risk patient (U1), it can specify a risk level that matches the signs of the predicted risk behavior among multiple risk levels. For example, as discussed above, assume that the signs of the predicted risk behavior of the at-risk patient (U1) are signs of risk behavior related to harm to others. The control unit (140) can specify the first risk level related to harm to others among multiple risk levels.

And, the control unit (140) can generate response action information (or action information, or preventive action information, or response information) to prevent self-harm or harm to others related to signs of risky behavior based on a specific level of risk.

In the present invention, the generated action information may be different for each of the multiple risk levels. Different response action information may be matched and exist for each of the multiple risk levels. Here, the different response action information may include first response action information matched to the first risk level and second action information matched to the second risk level.

For example, the different action information may include at least one of: i) information related to the deployment of personnel (e.g., security personnel, medical personnel, etc.) to prevent harm to others or self-harm, ii) time required for response to prevent harm to others or self-harm, iii) information for managing patients after completion of measures to prevent harm to others or self-harm (e.g., strengthening monitoring, improving the surrounding environment, notification information, etc.), iv) information on education and treatment programs to prevent harm to others or self-harm. However, the information included in the different action information is not necessarily limited to this, and various other measures may be included in addition to those mentioned.

At this time, the actions taken depending on the information contained in each of the first action information and the second action information may be different.

For example, in the case of the first action information, “5” people may be needed to prevent the at-risk patient (U1) from harming others. In contrast, in the case of the second action information, “2” people may be needed to prevent the at-risk patient (U1) from harming others, which is less than the number of people needed to prevent the at-risk patient (U1) from harming others.

As another example, for the first action information, “two” monitoring personnel may be needed to manage the at-risk patient (U1) after the completion of the prevention of other-injury accidents. In contrast, for the second action information, “one” monitoring personnel may be needed to manage the at-risk patient (U1) after the completion of the prevention of self-injury accidents.

As another example, for the first action information, information on education and treatment programs related to preventing harm to others may be included to prevent harm to others by the at-risk patient (U1). Alternatively, for the second action information, information on education and treatment programs related to preventing harm to others may be included to prevent harm to themselves by the at-risk patient (U1).

In this regard, the control unit (140) can generate first response measure information for preventing an accident involving another person if the specified risk level corresponds to the first risk level. Alternatively, if the specified risk level corresponds to a second risk level different from the first risk level, the control unit (140) can generate second response measure information for preventing a self-harm accident.

In addition, weights may be set for each of the multiple risk levels. In this case, the preset weights may be i) set by the self-harm and other-harm accident early prediction system (100) or ii) set by an administrator (e.g., medical staff) using the self-harm and other-harm accident early prediction system (100).

For example, if the weight is set by the self-harm and other-harm accident early prediction system (100) itself, the weight can be set based on whether the at-risk patient (U1) is more related to self-harm or other-harm accidents. For another example, if the weight is set by the manager using the self-harm and other-harm accident early prediction system (100), the weight can be set based on which accident is judged to require faster action, self-harm or other-harm accidents.

That is, by generating and providing action information according to the predicted risk level, the present invention can effectively prevent self-harm and harm to others by patients in medical institutions, and can contribute to customized management of patients and efficient management of medical resources.

Meanwhile, the control unit (140) can provide notification information for preventing self-harm or other-harm accidents to the control server (1000) linked to the early prediction system for other-harm accidents (100).

The control unit (140) can generate notification information including action information generated according to signs of predicted dangerous behavior, and provide the generated notification information to the control server (1000). For example, as illustrated in FIG. 6, the control unit (140) can output notification information including first action information for preventing a harm accident of a dangerous patient (U1) (ex: “As a harm accident is predicted in Zone A, the following actions are required”, “Target: Patient A”, “Required personnel: 1 doctor, 2 nurses, 2 security personnel”, “Response manual: Immediate isolation from target of harm required”, 600) to the display unit of the control server (1000).

At this time, as examined above, if a weight is set according to risk, the notification priority, notification size (e.g., notification window size, notification voice and vibration size), notification duration, notification repetition cycle, etc. of the notification information provided to the control server may be set and provided differently. As an example, if a higher weight is set for the first risk, the size of the notification information including the first action information may be provided larger than the size of the notification information including the second action information.

Meanwhile, at least one of the biometric information, behavioral information, and eye tracking information that may indicate risky behaviors related to self-harm or harm to others may differ from patient to patient.

For example, as illustrated in FIG. 7, a first patient (710) may be more likely to engage in risky behaviors related to self-harm when the pulse rate is “90 bpm” or higher, and a second patient (720) may be more likely to engage in risky behaviors related to self-harm when the pulse rate is “100 bpm” or higher.

For another example, patient 1 (710) may be more likely to exhibit risky behaviors associated with self-harm when sitting still and shaking in silence, and patient 2 (720) may be more likely to exhibit risky behaviors associated with self-harm when making fists with both hands.

That is, the degree of vital signs that indicate a risky behavior may vary for each patient (or at-risk patient), and such information may be stored and present in the early prediction system for other-cause accidents (100).

The control unit (140) can predict signs of risky behavior related to self-harm or harm to others differently for each patient by using an artificial intelligence model (150) learned from information on different patients. For example, as illustrated in FIG. 8, the control unit (140) can analyze an image (810) captured of a first patient (710) using an artificial intelligence model (150) to extract biometric information (ex: “ex: “Body temperature: 37.5°C”, “Pulse: 90 bpm”, “Blood pressure: 130/85 mmHg”, “Respiratory rate: 23 breaths per minute”, 811) of the first patient (710), behavioral information (ex: ex: “Sitting still and shivering in silence”, “Looking out the window and then turning the head to continuously search the surroundings”, 812) and gaze tracking information (ex: “Gaze time: 3 seconds or more”, “Gaze target: Outside the window”, “Gaze movement path: Looking out the window once and then looking at the surroundings”, “Gaze frequency: Repeatedly at 3 to 5 second intervals”, 813).

In addition, the control unit (140) can predict signs of risky behavior related to self-harm of the first patient (710) by using the extracted biometric information (811), behavioral information (812), and eye tracking information (813) of the first patient (710).

Furthermore, the control unit (140) can specify a second risk level based on the predicted signs of risky behavior related to self-harm accidents and generate second action information matching the second risk level. Then, the control unit (140) can generate notification information including the generated second action information and provide it to the control server (1000). For example, as illustrated in FIG. 9, the control unit (140) can output notification information including the second action information (ex: “Since a self-harm accident is predicted in Zone A, the following actions are required”, “Target: Patient 1”, “Required personnel: 1 doctor, 1 nurse”, “Response manual: Isolation from dangerous objects is required”, 900) to the display unit of the control server (1000).

As examined above, the artificial intelligence-based self-harm and other-harm accident early prediction method and system according to the present invention can contribute to preventing unnecessary accidents and efficiently managing the resources (e.g., manpower, time, cost, etc.) of medical institutions by early predicting risky behaviors related to self-harm or other-harm accidents.

In addition, the method and system for early prediction of self-harm and harm to others based on artificial intelligence according to the present invention can contribute to early prediction and prevention of risk behaviors related to self-harm or harm to others of at-risk patients and to establishing a medical service environment that is safe for patients and medical staff and helpful in increasing treatment effectiveness by building a data integration management platform with information sensed from various sensors and immediately applicable to the field.

Furthermore, the artificial intelligence-based method and system for early prediction of self-harm and harm to others according to the present invention can provide a medical service environment in which medical staff working at medical institutions can protect themselves from patients at high risk of harm to others by conducting a linkage analysis of the mutual relationship between at-risk patients and specific targets.

Furthermore, the artificial intelligence-based method and system for early prediction of self-harm and harm to others according to the present invention can enable the development of various digital therapeutics targeting self-harm and harm to others by constructing a database based on sensing data related to self-harm and harm to others.

Furthermore, the AI-based self-harm and other-harm accident early prediction method and system according to the present invention can early predict a patient's risk behavior related to self-harm or other-harm accidents and generate and provide response measure information that can proactively respond to the predicted risk behavior. Through this, medical institutions can drastically reduce the economic burden by reducing the demand for personnel such as safety personnel in the long term.

Meanwhile, the present invention discussed above can be implemented as a program that is executed by one or more processes on a computer and can be stored on a medium (or recording medium) that can be read by the computer.

Furthermore, the present invention discussed above can be implemented as a computer-readable code or command on a medium in which a program is recorded. That is, the present invention can be provided in the form of a program.

Meanwhile, computer-readable media include all types of recording devices that store data that can be read by a computer system. Examples of computer-readable media include hard disk drives (HDDs), solid state disks (SSDs), silicon disk drives (SDDs), ROMs, RAMs, CD-ROMs, magnetic tapes, floppy disks, and optical data storage devices.

Furthermore, the computer-readable medium may be a server or cloud storage that includes storage and that the electronic device can access through communication. In this case, the computer can download the program according to the present invention from the server or cloud storage through wired or wireless communication.

Furthermore, in the present invention, the computer described above is an electronic device equipped with a processor, i.e., a CPU (Central Processing Unit), and there is no particular limitation on its type.

Meanwhile, the above detailed description should not be construed as restrictive in all respects but should be considered as illustrative. The scope of the present invention should be determined by a reasonable interpretation of the appended claims, and all changes within the equivalent scope of the present invention are included in the scope of the present invention.

Claims (10)

A step of receiving sensing data on a risk patient sensed from at least one sensor;
A step of analyzing the sensing data using a learned artificial intelligence model; and
A step of predicting signs of risk behavior related to one of the signs of different risk behaviors related to self-harm or harm to others of the at-risk patient by using the analysis results through the artificial intelligence model;
A step of specifying a risk level according to a sign of the predicted risk behavior among multiple risk levels set according to the signs of the above different risk behaviors; and
A step of generating response action information to prevent any one of the above accidents related to the signs of the above predicted risk behavior based on the above specified risk level,
Signs of the different risk behaviors mentioned above are:
Including signs of risk behavior related to the above-mentioned harm incident and signs of risk behavior related to the above-mentioned self-harm incident,
The above multiple risks are:
The first risk level is determined based on the predicted signs of risk behavior related to the above-mentioned accidents, and
Includes a second risk level specified based on the predicted signs of risk behavior related to the above self-harm incident,
For each of the above multiple risks, a preset weight is set and exists.
In the step of generating the above response action information,
A method for early prediction of self-harm or harm to others, characterized in that it generates different response measure information for preventing self-harm or harm to others according to a specific risk level among a plurality of risk levels for which the preset weights are set. In the first paragraph,
The above sensing data is,
Contains the biometric information of the above-mentioned at-risk patient and images taken of the above-mentioned at-risk patient,
The above biometric information is,
A method for early prediction of self-harm or harm to others, characterized in that it includes at least one of body temperature, pulse, blood pressure, respiratory rate, and skin tension of the above-mentioned at-risk patient. In the second paragraph,
In the step of analyzing the above sensing data,
Using the above artificial intelligence model, the behavior of the at-risk patient is analyzed from the above image,
A method for early prediction of self-harm or harm to others, characterized by extracting behavioral information of a risk patient by using the analysis results of the behavior of the risk patient. In the third paragraph,
In the step of analyzing the above sensing data,
Using the above artificial intelligence model, the gaze and gaze patterns of the at-risk patient are analyzed from the above image,
A method for early prediction of self-harm or harm to others, characterized by extracting gaze tracking information of the at-risk patient by using the analysis results of the gaze and gaze patterns of the at-risk patient. In paragraph 4,
The above gaze tracking information is,
A method for early prediction of self-harm or harm to others, characterized by including at least one of the gaze movement path, gaze target, gaze time, and gaze frequency of the above-mentioned at-risk patient. In paragraph 4,
Signs of the above risk behaviors include:
A method for early prediction of self-harm or harm to others, characterized in that the prediction is made using the biometric information, the behavioral information, and the eye tracking information of the at-risk patient. In the first paragraph,
The above multiple risks are:
Including the first risk level and the second risk level specified from the signs of the above different risk behaviors,
The first risk above is,
Specified based on the prediction of signs of risk behavior related to the above-mentioned accident through the above-mentioned artificial intelligence model,
The second risk level above is,
A method for early prediction of self-harm or harm to others, characterized in that the signs of risk behavior related to the self-harm incident are predicted through the artificial intelligence model. In Article 7,
For each of the above multiple risk levels, different response action information is matched and exists.
The above different response measures information is:
Including first response measure information matched to the first risk level and second response measure information matched to the second risk level,
In the step of generating the above action information,
If the above-mentioned specific risk level corresponds to the above-mentioned first risk level, the above-mentioned first response measure information is generated to prevent the above-mentioned third-party accident,
A method for early prediction of self-harm or harm to others, characterized in that when the above-mentioned specific risk level corresponds to the second risk level different from the above-mentioned first risk level, the second response measure information for preventing the self-harm accident is generated. In an early prediction system for self-harm or harm to others, including a communication unit, a storage unit, a control unit, and an artificial intelligence model,
The above system,
Receive sensing data about a risk patient sensed from at least one sensor,
Using the above learned artificial intelligence model, the sensing data is analyzed,
Using the analysis results through the artificial intelligence model, predict the signs of risk behavior related to one of the signs of risk behavior related to self-harm or harm to others of the at-risk patient,
Among the multiple risk levels set according to the signs of the above different risk behaviors, the risk level according to the signs of the predicted risk behavior is specified.
Based on the above specified risk level, generate response action information to prevent any of the above accidents related to the signs of the above predicted risk behavior,
Signs of the different risk behaviors mentioned above are:
Including signs of risk behavior related to the above-mentioned harm incident and signs of risk behavior related to the above-mentioned self-harm incident,
The above multiple risks are:
The first risk level is determined based on the predicted signs of risk behavior related to the above-mentioned accidents, and
Includes a second risk level specified based on the predicted signs of risk behavior related to the above self-harm incident,
For each of the above multiple risks, a preset weight is set and exists.
The above control unit,
An early prediction system for self-harm or harm to others, characterized in that it generates different response measure information for preventing self-harm or harm to others according to a specific risk level among a plurality of risk levels for which the preset weights are set. A program that is executed by one or more processes on an electronic device and stored on a computer-readable recording medium.
The above program is,
A step of receiving sensing data on a risk patient sensed from at least one sensor;
A step of analyzing the sensing data using a learned artificial intelligence model; and
A step of predicting signs of risk behavior related to one of the signs of different risk behaviors related to self-harm or harm to others of the at-risk patient by using the analysis results through the artificial intelligence model;
A step of specifying a risk level according to a sign of the predicted risk behavior among multiple risk levels set according to the signs of the above different risk behaviors; and
Includes commands for performing a step of generating response action information for preventing any one of the above accidents related to the signs of the above predicted risk behavior based on the above specified risk level,
Signs of the different risk behaviors mentioned above are:
Including signs of risk behavior related to the above-mentioned harm incident and signs of risk behavior related to the above-mentioned self-harm incident,
The above multiple risks are:
The first risk level is determined based on the predicted signs of risk behavior related to the above-mentioned accidents, and
Includes a second risk level specified based on the predicted signs of risk behavior related to the above self-harm incident,
For each of the above multiple risks, a preset weight is set and exists.
In the step of generating the above response action information,
A program stored in a computer-readable recording medium, characterized in that it generates different response measure information for preventing self-harm or harm to others according to a specific risk level among a plurality of risk levels for which the preset weights are set.

Images