KR102140585B1 - Apparatus for generating rules of event calculus based on ontoloby for activity intension recognition and method thereof - Google Patents

Abstract

Disclosed is an apparatus and method for generating an ontology-based event operation rule for recognizing a person's intention to act according to an embodiment. The apparatus for generating an event operation rule may include a fluent mapper that extracts contents necessary for reasoning of behavioral intention from among contents included in the behavioral intention ontology and converts it into a predefined fluent; An event recognizer that converts percept information about a person's life pattern into predefined event information; And a rule converter that generates an event operation rule for recognizing an intention to act based on the converted fluent and the converted event information.

Description

A device and method for generating an ontology-based event calculation rule for recognizing a person's behavioral intention {APPARATUS FOR GENERATING RULES OF EVENT CALCULUS BASED ON ONTOLOBY FOR ACTIVITY INTENSION RECOGNITION AND METHOD THEREOF}

Embodiments relate to a behavior intention recognition technology, and more particularly, to an ontology-based event operation rule generating apparatus and method for recognizing a person's behavior intention.

The method of inferring behavior or intention through universal knowledge-based rules has the advantage that training data is not required to train the model. For this reason, techniques for recognizing the intention of action by applying the event operation have been published.

However, in the prior art, the rules for calculating an event were directly defined in order to construct a combination of long-term actions based on the recognition of the intention to act. Therefore, the applied event operation rules reflected the knowledge of the individual developers who built the system rather than expert information.

Registered Patent Publication No. 10-1766583

An embodiment may provide an ontology-based event calculation rule generating apparatus and method for recognizing a person's intention to act.

An apparatus for generating an event operation rule according to an embodiment of the present invention includes: a flexible mapper that converts contents necessary for reasoning of behavioral intention from among contents included in the behavioral intention ontology into a predefined fluent; An event recognizer that converts percept information about a person's life pattern into predefined event information; And a rule converter that generates an event operation rule for recognizing an intention to act based on the converted fluent and the converted event information.

The fluent mapper includes a definition unit that defines a fluent that is applied to the event calculation rule; An extracting unit for extracting contents necessary for reasoning of the intention of the behavior among contents included in the ontology of the behavior intention; And it may include a conversion unit for converting the extracted content to the defined fluent.

The predefined fluent may include a behavior intention fluent representing a behavioral intention change, a behavioral fluent representing a behavioral change, a posture fluent representing a posture change, and a peripheral object fluent representing a surrounding object change. have.

The event recognizer defines a unit defining event information applied to the event operation rule; An input unit for receiving percept information in which information on the person's life pattern is defined in time series; And it may include a conversion unit for converting the input percept information to the defined event information.

The predefined event information may include a behavior event for a person's behavior change, a posture event for a person's current posture, and a surrounding object event for changing a state of the surrounding object.

The rule converter combines the converted fluent and the converted event information into a relationship between a behavioral intention and an action, a posture, and surrounding objects for content necessary for inferring the behavioral intention, and sets the event calculation rule as a result of the combination. Can be created.

A method for generating an event operation rule according to another embodiment of the present invention includes converting a content necessary for inferring an action intention from among content included in an action intention ontology into a predefined fluent; Converting percept information about a person's life pattern into predefined event information; And generating an event calculation rule for recognizing an intention to act based on the converted fluent and the converted event information.

The converting to the fluent may include defining a fluent applied to the event calculation rule; Extracting the content necessary for reasoning of the intention of the action among the content included in the ontology of the action intention; And converting the extracted contents into the defined fluent.

The converting to the event may include defining event information applied to the event calculation rule; Receiving the percept information in which information about the person's life pattern is defined in time series; And converting the input percept information into the defined event information.

In the generating step, the event calculation rule is calculated as a result of combining the converted fluent and the converted event information in a relationship between action intention and action, posture, and surrounding objects for the content necessary for inferring the action intention. Can be created.

According to an embodiment, the content included in the behavioral intention ontology is converted into a fluent, the percept information on a person's life pattern is converted into event information, and then the behavioral intention is recognized based on the converted fluent and the event information. For example, by generating a rule for calculating an event, it may be possible to apply expert knowledge for recognition of an intention to act more easily.

According to the embodiment, since it is possible to apply expert knowledge for the recognition of the intention to act, various systems for healthcare can be developed more efficiently.

1 is a diagram illustrating an apparatus for generating an event operation rule according to an embodiment of the present invention.
FIG. 2 is a diagram showing a detailed configuration of the fluent mapper shown in FIG. 1.
FIG. 3 is a diagram showing a detailed configuration of the event recognizer shown in FIG. 1.
4 is a diagram illustrating a method for generating an event operation rule according to an embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

However, the technical spirit of the present invention is not limited to some embodiments described, but may be implemented in various different forms, and within the technical spirit scope of the present invention, one or more of its components between embodiments may be selectively selected. It can be used by bonding and substitution.

In addition, terms (including technical and scientific terms) used in the embodiments of the present invention, unless specifically defined and described, can be generally understood by those skilled in the art to which the present invention pertains. It can be interpreted as meaning, and commonly used terms, such as predefined terms, may interpret the meaning in consideration of the contextual meaning of the related technology.

In addition, the terms used in the embodiments of the present invention are for describing the embodiments and are not intended to limit the present invention.

In the present specification, the singular form may also include the plural form unless specifically stated in the phrase, and is combined with A, B, C when described as “at least one (or more than one) of A and B, C”. It can contain one or more of all possible combinations.

In addition, in describing the components of the embodiments of the present invention, terms such as first, second, A, B, (a), and (b) may be used.

These terms are only for distinguishing the component from other components, and the term is not limited to the nature, order, or order of the component.

And, when a component is described as being'connected','coupled' or'connected' to another component, the component is not only directly connected, coupled or connected to the other component, but also to the component It may also include the case of'connected','coupled' or'connected' due to another component between the other components.

In addition, when described as being formed or disposed in the “upper (upper) or lower (lower)” of each component, the upper (upper) or lower (lower) is one as well as when the two components are in direct contact with each other. It also includes a case in which another component described above is formed or disposed between two components. In addition, when expressed as “up (up) or down (down)”, it may include the meaning of the downward direction as well as the upward direction based on one component.

In an embodiment, the content included in the behavioral intention ontology is converted into a fluent, the percept information on a person's life pattern is converted into event information, and then the behavioral intention is recognized based on the converted fluent and event information. We propose a new method to create an event operation rule.

At this time, the fluent (fluent) is defined as a term 1 term or predicate whose value changes with time.

1 is a diagram illustrating an apparatus for generating an event operation rule according to an embodiment of the present invention.

Referring to FIG. 1, an apparatus for generating an event operation rule according to an embodiment of the present invention may include a fluent mapper 100, an event recognizer 200, a rule converter 300, and an event operation knowledge base 400. have.

The flexible mapper 100 may convert content necessary for reasoning of the intention among the content included in the behavioral intention ontology into fluent applied to a predefined event operation rule. Here, the behavioral intention ontology defines the relationship between a person's behavioral intention and a person's behavior, posture, and surrounding objects in a format that can share expert knowledge.

FIG. 2 is a diagram showing a detailed configuration of the fluent mapper shown in FIG. 1.

Referring to FIG. 2, the fluent mapper 100 according to an embodiment of the present invention converts content related to intention inference into a fluent, and defines, 110, extracts 120, and converts 130 It may include.

The definition unit 110 may define a fluent form applied to an event operation rule. The form of the fluent is defined as four types such as a behavior intent fluent, a behavioral fluent, a posture fluent, and a surrounding object fluent, and can be expressed as the following [Table 1].

division Explanation Expression form Intention fluent Expressing a change in behavior intention(p, i) Action fluent Express behavior change action(p, s) Change of pose fluent Express posture change pose(p, s) With object fluent Expresses changes in surrounding objects With(p, o, on/off)

In [Table 1], intention fluent is expressed as intention (person_id, intention_id). For example, in the case of intention (kim, meal), the person named'kim' has an intention of'meal' Express. The action fluent expresses the action change such as'the person picked up the spoon' as an action(kim, pickup_spoon), and the change of pose fluent poses the same as'the person stands' The change is expressed in pose(kim, stand), and the surrounding object fluent is with(kim, tv, on) the change in the object around the person or the body such as'the surrounding TV is on'. It is expressed as

The extraction unit 120 may extract content necessary for reasoning of the intention of the action from among contents included in the ontology of the intention to act.

At this time, the form for extracting the contents necessary for inferring the behavioral intention can be expressed as [Table 2] below.

Class(Intention),
subClassOf(?x, Intention), Class(?x),
Class(Person), Class(Action), Class(Pose), Class(Object)
Action(?y), Pose(?z), Object(?j)
InstanceOf(?x, ?i001)
composedAction(?i001, ?y), composedPose(?i001, ?z)
composedObjects(?i001, ?z), isTurnedOn(?z, ?b)

In [Table 2], the symbol with'?' means a variable, and indicates that all predicates matching the rest of the predicate except for the part in which the variable is located are extracted.

An example of extracting the content necessary for inferring behavioral intention using the form of [Table 2] is as shown in [Table 3] below.

Class(Intention),
subClassOf(Meal, Intention), Class(Meal), ...
Class(Person), Class(Action), Class(Pose), Class(Object), ...
Action(pickup_spoon), Action(drink_liquid), Action(use_vacuum), Action(use_remote), ...
Pose(sitOnChair), Pose(sitOnFloor), Pose(stand), ...
Object(table), Object(spoon), Object(tv), ...
InstanceOf(Meal, mealFluent001)
composedAction(mealFluent001, pickup_spoon),
composedPose(mealFluent001, sitOnChair)
composedObjects(mealFluent001, table), isTurnedOn(table, false)
...

In [Table 3] above, the action intent'meal', action'pickup_spoon','drink_liquid','use_vacum','use_remote', posture'sitOnChair','sitOnFloor','stand', surrounding objects'table', ' It represents extracting spoon' and'tv'.

The conversion unit 130 may convert the content necessary for inferring the extracted behavioral intention into a predefined fluent. At this time, an example of converting to a predefined fluent is shown in [Table 4] below.

Conversion target Fluent transformed example Subclass of all intention classes Intention fluent Class(Meal) -> intention(?p, Meal) Instance name of all action classes Action fluent Action(pickup_spoon) -> action(?p, pickup_spoon) Instance name of all pose classes Change of pose fluent Pose(sitOnChair) -> pose(?p, sitOnChair) Instance name of all object classes With object fluent Object(table) -> with(?p, table, on/off)

In [Table 4], when the content of the ontology is converted into a fluent,'?p', which is person_id, is expressed as a variable to indicate based on the person who is the center of the percept input in real time for recognition of the intention of action. . Therefore, the Class(Meal) predicate, which is a class for Meal intention, can be converted into an intention (?p, Meal), which is a behavioral intention fluent meaning that any person'?p' has Meal intention.

That is, the behavioral intention''meal'' is converted to the behavioral intention fluent, the behavior'pickup_spoon' is converted to the behavioral fluent, the posture'sitOnChair' is converted to the posture fluence, and the surrounding object'table' is the surrounding object fluence. Represents conversion to unt.

The event recognizer 200 may convert percept information into predefined event information. Here, the percept information generates a life pattern within a person's work and time as a probability model through a Gaussian mixture model, and then generates intentions, actions, and attitudes generated based on the model in a time series of seconds. Data.

FIG. 3 is a diagram showing a detailed configuration of the event recognizer shown in FIG. 1.

Referring to FIG. 3, the event recognizer 200 according to an embodiment of the present invention converts percept information into event information, and may include a definition unit 210, an input unit 220, and a conversion unit 230 have.

The definition unit 210 may define event information for inferring an action intention. Such event information is defined in three types such as action event, posture event, and surrounding object event, and can be expressed as the following [Table 5].

division Explanation Expression form action event Events about human behavior changes Called_action(p) pose event Events about a person's current posture Called_pose(p) sense object event Events that change the state of surrounding objects status(p, o, on/off)

In [Table 5], the events are classified into three types, and based on percept information input according to each expression type, the events, postures, and states of surrounding objects generated at the present time are expressed.

The input unit 220 may receive percept information in which information about a person's life pattern of a person is defined in time series.

The converter 230 may convert the input percept information into predefined event information.

For example, in the converter 230, the current situation that a person named'kim' is sitting with a spoon and the surrounding objects are a spoon and a table is {kim, pickup_spoon, sitOnChair, {[spoon, off], [table, off]}} If entered as a percept expressed as, event (called_pose about the attitude that the person named'kim' is sitting (pickup_spoon(kim) where called_action is pickup_spoon pickup_spoon) and person of'kim' is sitting. The sitOnChair, sitOnChair(kim)), and the person's surrounding object called'kim' are converted into events (status(kim, spoon, off)) about the surrounding object called spoon.

The rule converter 300 may generate an event calculation rule for recognizing an intention of action based on the fluent transformed by the fluent mapper 100 and the event information transformed by the event recognizer 200.

The three rule frames IR0, IR1, and IR2 that generate these event operation rules are defined as in [Table 6] below.

IR0: Happens(e, t) :: Happens(en(p), t), Happens(status(p, o, on/off), t),
en: event predicate
IR1: Initiates(e, f, t) :: Initiates(en(p), action(p, sn), t), en=sn.
Initiates(en(p), pose(p, sn), t),
en=sn, sn: event_id
Initiates(status(p, o, on/off), with(p, o, on/off), t).
IR2: HoldsAt(f1, t) ∧ HoldsAt(f2, t) ∧ HoldsAt(f3, t) => Initiates(e, f, t)

In [Table 6], various event calculation rules are required to recognize the intention of action, and in order to automatically configure them, the fluent mapper and event recognizer automatically generated in the fluent mapper and event recognizer according to the three basic frames specified above. Rules for recognizing behavioral intention are automatically generated by combining the contents extracted from the behavioral intention ontology in relation to the behavioral intention. Here, the relationship of the latitude of the row means the relationship between the intention of action and the action, posture, and surrounding objects.

Here, IR0 is generated whenever percept information is input, and Initiates (e, f, t) constituting IR1 and IR2 are intention class objects and intention class objects based on the results extracted for intention of action from ontology in the fluent mapper. It is automatically generated at the same time by referring to the predicate relationship with composedAction, composedPose, composedObjects, and isTurnedOn.

At this time, an example of generating an event operation rule is shown in the following [Table 7] and [Table 8].

InstanceOf(Meal, mealFluent001)
composedAction(mealFluent001, pickup_spoon),
composedPose(mealFluent001, sitOnChair)
composedObjects(mealFluent001, table), isTurnedOn(table, false)
...

Happen(pickup_spoon(p), t), Happens(sitOnChair(p), t), Happens(status(p, table, off), t)
Initiates(pickup_spoon(p), action(p, pickup_spoon), t)
Initiates(sitOnChair(p), pose(p, sitOnChair), t)
Initiates(status(p, table, off), with(p, table, off), t)
· HoldsAt(action(p, pickup_spoon), t) ∧ HoldsAt(pose(p, sitOnChair), t) ∧
HoldsAt(with(p, table, off), t) => Initiates(pickup_spoon(p), intention(p, Meal), t)

Events that are automatically generated according to the defined format (pickup_spoon(kim), sitOnChair(kim), status(kim, spoon, off), etc.) are Happens(e, e, such as Happens(pickup_spoon(kim), t) around the time t t) It is generated in IR0 format. When IR0 is performed, IR1 makes action fluent and change of pose fluent and with object fluent corresponding to it based on the relationship between the behavioral intention extracted from the behavioral intention ontology, and applies it to IR2. pickup_spoon(p), intention(p, Meal), and t) are generated. The event operation knowledge base 400 may store the generated event operation rules, fluents, and event information.

4 is a diagram illustrating a method for generating an event operation rule according to an embodiment of the present invention.

Referring to FIG. 4, the apparatus for generating an event operation rule according to an embodiment of the present invention extracts content necessary for inferring an action intention from among content included in an intention to act (S410 ), and extracts content related to the extracted intention inference. It can be converted to a predefined fluent (S420).

Next, the apparatus for generating an event operation rule may convert percept information from a user into predefined event information (S430).

Next, the apparatus for generating an event calculation rule may generate an event calculation rule for recognizing an intention of action based on the converted fluent and the converted event information (S440). In this case, the apparatus for generating an event operation rule may generate the event operation rule as a result of combining the converted fluent and the converted event information in a relationship of intention to act on the extracted content.

The term'~ unit' used in this embodiment refers to hardware components such as software or field-programmable gate array (FPGA) or ASIC, and'~ unit' performs certain roles. However,'~ wealth' is not limited to software or hardware. The'~ unit' may be configured to be in an addressable storage medium or may be configured to reproduce one or more processors. Thus, as an example,'~ unit' refers to components such as software components, object-oriented software components, class components, and task components, processes, functions, attributes, and procedures. , Subroutines, segments of program code, drivers, firmware, microcode, circuitry, data, database, data structures, tables, arrays, and variables. The functions provided within components and'~units' may be combined into a smaller number of components and'~units', or further separated into additional components and'~units'. In addition, the components and'~ unit' may be implemented to play one or more CPUs in the device or secure multimedia card.

Although described above with reference to preferred embodiments of the present invention, those skilled in the art variously modify and change the present invention without departing from the spirit and scope of the present invention as set forth in the claims below. You can understand that you can.

100: fluent mapper
110: justice
120: extraction unit
130: conversion unit
200: event recognizer
210: Justice
220: input
230: converter
300: rule converter
400: case operation knowledge base

Claims (10)

A Fluent Mapper that converts the content included in the behavioral intention ontology into a predefined fluent;
An event recognizer that converts percept information about a person's life pattern into predefined event information; And
And a rule converter for generating an event calculation rule for recognizing an intention to act based on the converted fluent and the converted event information,
The predefined fluent includes a behavior intention fluent representing a behavioral intention change, a behavioral fluent representing a behavioral change, a posture fluent representing a posture change, and a peripheral object fluent representing a surrounding object change,
The predefined event information includes a behavior event for a person's behavior change, a posture event for a person's current posture, and a surrounding object event for changing a state of the surrounding object,
The rule converter combines the converted fluent and the converted event information into a relationship between a behavioral intention and an action, a posture, and surrounding objects for content necessary for inferring the behavioral intention, and sets the event calculation rule as a result of the combination. A device for generating and calculating event rules. According to claim 1,
The fluent mapper,
A definition unit defining a fluent applied to the event calculation rule;
An extracting unit for extracting contents necessary for reasoning of the intention of the behavior among contents included in the ontology of the behavior intention; And
And a conversion unit converting the extracted contents into the defined fluent. delete According to claim 1,
The event recognizer,
A definition unit defining event information applied to the event operation rule;
An input unit for receiving the percept information in which information about the person's life pattern is defined in time series; And
And a conversion unit configured to convert the input percept information into the defined event information. delete delete Converting the content included in the behavioral intention ontology into a predefined fluent;
Converting percept information about a person's life pattern into predefined event information; And
Generating an event operation rule for recognizing an intention to act based on the converted fluent and the converted event information,
The predefined fluent includes a behavior intention fluent representing a behavioral intention change, a behavioral fluent representing a behavioral change, a posture fluent representing a posture change, and a peripheral object fluent representing a surrounding object change,
The predefined event information includes a behavior event for a person's behavior change, a posture event for a person's current posture, and a surrounding object event for changing a state of the surrounding object,
In the generating step, the event calculation rule is the result of combining the converted fluent and the converted event information in relation to the action intention and action, posture, and surrounding objects for the content necessary for inferring the action intention. A method of generating an event operation rule. The method of claim 7,
The step of converting to the fluent,
Defining a fluant applied to the event operation rule;
Extracting the content necessary for reasoning of the intention of the action among the content included in the ontology of the action intention; And
And converting the extracted content into the defined fluent. The method of claim 7,
The step of converting to the event,
Defining event information applied to the event operation rule;
Receiving the percept information in which information about the person's life pattern is defined in time series; And
And converting the input percept information into the defined event information. delete

Images