A PERCEPTION MECHANISM FOR TWO-DIMENSIONAL

SHAPES IN THE VIRTUAL WORLD

Jae-Woo Park and Jong-Hee Park

Kyungpook Nat’l University, Daegu, South Korea

Keywords: Perception, Ontology, Recognition.

Abstract: Lifelike agent in the virtual world is an agent who is designed to be able to simulate the realistic human

behavior. Agents continuously repeat the process that includes perception, recognition, decision and

behavior in the virtual world. Through those processes, the agents store new information in their memory or

modify their knowledge if it is needed. This study mainly deals with the perception that is intermediate step

between image processing and recognition. In this study, you will see how the agents perceive shapes. And

you also will realize how it is possible to infer the part of shape that was partially hidden from the agent’s

vision.

1 INTRODUCTION

We need to implement some kind of functions such

as perception, recognition and behavior in order to

make a system that can operate like real human in

the virtual world, and we have to reorganize these

functions smoothly. The fact that sees through the

eye is natural process. However, if you analyze the

process of perception, you could see many things

that have been occurred progressively for a short

time. Image processing is the starting point of the

agent's recognition. It is called image processing

which convert an image to digital data. The

perception which generated by Image processing

using points, lines and colors of pixels is to identify

two-dimensional shape. The rule-base system is used

to divide the shapes that touch each other. It is also

used to infer the shapes which are partially hidden

by other shapes or are partially outside the visions of

agents. Main purpose of perception is to increase the

success rate of recognition. In chapter 3.1, we

explain how to model the agent’s vision. In chapter

3.2, we explain the perception’s primitive units such

as points, lines and colors of pixels. The chapter 3.3

explains the inference of shape. In chapter 4, we

introduce spatial relation of shapes which are

contacted each other. Final chapter presents

requirements for successful recognition.

2 RELATED WORK

2.1 Virtual World

The ultimate aim of the Virtual World project is to

develop an authentic simulated cosmos which

consists of an authentic cyber space (Park, 2000) and

Virtual Inhabitants (SeJin and Kim and Park, 2004).

The authentic cyber space is a virtual environment

governed by the various rules and principles in a

systematic and organic manner (Park, 2000). The

Virtual Inhabitant is a computer-controlled agent

who dwells in the authentic cyber space and

interacts with the user in a manner similar to the

interaction among typical humans (SeJin, 2007).

2.2 Agent Model

An agent is a computer system that is situated in

some environment, and that is capable of

autonomous action in this environment in order to

meet its design objectives (Wooldridge, 2002).

Believable agent is personality-rich autonomous

agent with powerful properties of characters from

the arts (Bryan Loyall, 1997). Lifelike agent(or

lifelike character)is an agent which is designed in

order to simulate the realistic human(or animal)

behavior. These two agent architectures have very

similar requirements such as personality, emotion

and sociality because their ultimate goal is to induce

381

Park J. and Park J. (2010).

A PERCEPTION MECHANISM FOR TWO-DIMENSIONAL SHAPES IN THE VIRTUAL WORLD.

In Proceedings of the 12th International Conference on Enterprise Information Systems - Artiﬁcial Intelligence and Decision Support Systems, pages

381-384

DOI: 10.5220/0002869803810384

 SciTePress

user's interest and retain it. In addition, both

believable agent and lifelike agent are designed for

multimedia entertainment applications such as

computer games, computer animations, and virtual

communities and multimedia pedagogical

applications such as intelligent tutoring systems, and

simulated training systems (SeJin, 2007).

3 AGENT’S PERCEPTION

METHOD FOR

TWO-DIMENSIONAL SHAPES

3.1 Modelling of Vision

The agent has limited range of vision and the

perceivable information depends on the distance

from the agent to the target object. For a target

object O

Scale(O

) = f(Distance, Vision) where

Distance denotes the distance between the agent and

and Vision denotes the visual capability of the

agent. Vision determines the distance within which

the agent can discern an object. The vision V

(t) of

agent A is modeled by a cone with its vertex at her

eyes as illustrated in Figure 1. Its dimension is

specified with three parameters, i.e., the angle θ,

length l and direction d. While the first two

parameters depend on her bodily conditions, the

direction changes to her decision. Specifically, V

(t)

= (θ(t

), l(t

), d(t

+t)) where t

denotes the time with

a time unit in which bodily conditions may change.

An object within the cone denoted by the shaded

area is considered as perceived by the agent. The

cone will disappear, i.e., θ or l = 0, if her eyes are

closed or impaired. The l may be extended with help

of a device like a telescope or microscope. While +l

denotes the range of a natural vision or vision

extended with a telescope, -l denotes the range of a

microscopic vision. A dotted line indicates an

extended vision in either way. The object size would

be perceived by the agent as multiplied by a factor

proportional to |l|/d in case of +l and d/|l| in case of -l.

(J. H. Park., 2004)

Figure 1: Scope of vision.

3.2 Primitive Units for Perception

When we put a great effort to get something into our

friend’s head, it is good to explain typical features of

the object. This study defines the features as

characteristic property. For example, if we describe

a snowman to our friend, the snowman’s

characteristic properties are both one circle above

another circle and its color is white. In this study, the

characteristic properties of the objects to define them

are type of shapes, their spatial relations and their

colors. Figure 2 shows an object came within vision

of agent. The inside of circular area represents vision

of an agent. Our image processing step converts an

image in the scope into lines and points. The rapidly

changing RGB values from the left image of Figure

2 are used to identify the boundary lines. The right

image of Figure 2 shows the identified boundary

lines.

Figure 2: Image processing for object.

The points at the junctions of edges and the lines

occurring by RGB difference between the pixels are

used as the primitive units for two-dimensional

perception. The connected status, angle and colors

are also used to determine the shape type. These

parameters allow us to judge whether an object is

regular shape or not. Figure 3 shows detailed

primitive shape’s information. In Figure 3, point a, b

and c show the points indicating a junction. The

rectangle 2Lv2 has a section between Junctions

point a to point c (a-b-c). The number next to a line

shows its Draw value indicating how many times

used it.

Figure 3: Detailed information of primitive shape.

ICEIS 2010 - 12th International Conference on Enterprise Information Systems

382

For example, Line-ad is used in the rectangle-2Lv2

only once, and the Line-bc is used twice in the

rectangle-2Lv2 and the rectangle-2Lv7, their Draw

value being one & two respectively. Two connecting

lines at the Connection point have an angle between

them, gauged by the trigonometric functions.

3.3 Inference of Shape

When an agent perceives a target object that is

partially blocked by other objects or partially out of

the vision, she would use inference to complete her

perception. Figure 4 shows the information on the

Boundary lines for the shape being inferred (red

boundary of ID: 2LV7 shape). In the information,

we can find two sections (the red rectangle) on the

rapidly changing line’s length (L) and line’s angle

(A). The agent reforms her inferring only when the

target shape is not a regular shape, and its available

amount of information is to exceed minimum. An

agent excludes those rapidly changing parts and

starts her inferring with the rest of the information

on the image to find its closed regular shapes. And

the agent draws virtual lines starting from Junction

points. If it is judged to be a regular shape based on

the virtual lines, she adds the information on the

virtual lines to her original information.

Figure 4: Inferable shape and its Boundary Lines.

4 SPATIAL RELATIONS AMONG

SHAPES

With the shape type identified before, we now

identify the spatial relations existent among those

shapes to achieve a complete recognition. Defined

spatial relation in perception is different with the

recognition’s relation. The spatial relations as

perceived could be specialized into to diverse spatial

relations in recognition. We’ll follow the relations

set of HSR for our recognition (Choe and Park 2005).

For example, relations of HSR include among others

associated with ‘Touch’ in perception, ‘straddling’,

‘hanging’, ‘leaning’. In short, the relations of

perception are redefined again.

4.1 Connection among Shapes

If there is a connection among shapes, shape-A

based relation of shape-B is expressed in the

database on the ontology.

Table 1: Types of Connection.

Connection Defined rule

In The case when central point of B insides

boundaries of A. Only, The inference of A

does not occur by B

Out The case when central point of B outside

boundaries of A. Only, The inference of A

does not occur by B

Touch The case when B and A have shared lines

of the above one. Only, The inference does

not occur by shared lines.

Behind The case when B and A have shared lines

of the above one. And the inference occurs

by shared lines. In this case, The

connection of shape A is Touch and shape

B’ connection is Behind

The Out relation occurs only when the Out Touch

has been found. We mention only because there are

numerous Out relations including the Out touch.

From Table 1, the central point of shapes, shape’s

boundaries and their inference are determinant of

connection between them. These connections are

designed in the ontology and are stored in database.

When there are some additional connections (i.e.

repetition, symmetry and reverse etc.), this data

structure can be used to solve the problem by adding

proper rules in the ontology.

4.2 Direction among Shapes

The Direction is another parameter to special the

spatial relations. The Direction uses a shape’s

central point as the reference. For example, ‘direct3

out touch', means a shape has an Out relation on the

3:00 direction with another shape. Figure 5 shows

the coordinate for area. Central circle is called

Center relation. Other Coordinates proceed in order

clockwise. Various spatial relations could be defined

indicates by combining Connection and Direction.

A PERCEPTION MECHANISM FOR TWO-DIMENSIONAL SHAPES IN THE VIRTUAL WORLD

383

Figure 5: The Direction among shapes.

5 CONCLUSIONS

In this study, we introduce a perception mechanism

for the shapes that touch each other. We also

introduce the inference of shape and the spatial

relations among the component shapes of a

composite shape for recognizing objects in the

knowledge schema. The virtual world in general is

composed with regular shape. So, applying the

introduced methods to the virtual world is not so

difficult. However, in order to apply this technology

to a realistic virtual world, we need to implement the

superior image computing ability and to define

abstraction for a lot of objects. From now on, the

study on the perception mechanism will be focused

the irregular shape expressions and additional

relations among shapes.

REFERENCE

Park, J., 2000. “A platform for multimedia simulation of

spatio-temporal situations for declarative knowledge

learning”. AIMM Laboratory, Kyungpook National

University, Daegu, Korea, Tech. Rep.

Patrick, H, Winston. 1993. “Artificial intelligence”.

Addison-Wesley Publishing Company. pp. 81-94.,

Addison Wesley., 3

edition.

Park, J., 2004. “space”. AIMM Laboratory, Kyungpook

National University, Daegu, South Korea, Tech. Rep

14 p.10

SeJin, J., Kim, B., Park, J., 2004. “An agent architecture

for implementing a virtual Inhabitant”. In 10

Multimedia Modeling Conference, pp.316-322.

Bryan Loyall, A. 1997. “Believable agents: Building

interactive personalities”. Ph.D. dissertation Carnegie

Mellon University, Pittsburgh, PA,

Wooldridge, M., 2002. “An Introduction to Multi Agent

Systems”. John Wiley & Sons, Chichester, UK,

SeJin, J., 2007. “A Knowledge Model for Simulating

Human-like Behavior of Virtual Inhabitant”. Ph.D.

dissertation Kyungpook National University, Daegu,

South Korea

Park , J., 2000. “A logical simulation of spatio temporal

situations”, AIMM Laboratory, Kyungpook National

University, Daegu, Korea, Tech. Rep.,

S. H. Choe., J. H. Park, J., 2005. “Hierarchical spatial

relation based on a contiguity graph”. International

Journal of Intelligent Systems Vol.20, No. 9, pp867-

892

ICEIS 2010 - 12th International Conference on Enterprise Information Systems

384