[其它] stAR: Visualizing Constellations with Star Retrieval

彬彬

797 主题	1 听众	1万积分

资深设计师

Rank: 7 Rank: 7 Rank: 7

纳金币: 5568
精华: 0

电梯直达

楼主

发表于 2012-1-3 10:53:36 |只看该作者 |倒序浏览

1 Introduction

Thanks to the great feature that allows to intuitively visualize virtual

contents by superimposing them on real scenes, augmented

reality (AR) technologies have widely been used in various fields

such as entertainment, advertisement, education, tourism, and industrial/

medical applications. In particular, AR applications in education

have provided good solutions for attracting interests of users

and enhancing their understanding in learning capabilities. For example,

AR-based books are very helpful for users to easily understand

text-based knowledge by augmenting supplementary contents

on the books.

This paper presents a novel application for learning constellations

with AR. Our application intuitively provides constellation information

such as star pattern, boundary, and mythological illustration

to users by directly overlaying it on a celestial map. By utilizing

stars extracted from the map instead of visual markers, it also supports

immersive environments for constellation learning; thus, users

can seamlessly interact with visual contents. To develop the application,

we newly propose an approach that retrieves a star from a

star database with local geometrical links of its neighbor stars, described

by locally likely arrangement hashing (LLAH) [Nakai et al.

2006]. Our work was originally motivated for developing an educational

application for learning constellations, but promisingly, it

would be applicable to entertainments, games, and arts.

2 Methodology

For visualizing constellations on a celestial map, the proposed application

utilizes stars extracted from a target scene (celestial map),

without attaching any visual markers. Because the target scene has

a untextured background (sky) and dots (stars), moreover, conventional

descriptor-based approaches that use local features on a scene

are not feasible for matching stars. Therefore, the contributions in

our application are that a star itself is utilized as a visual marker

based on its geometrical arrangement of local neighborhood, and

each star is matched and identified based on a novel approach,

LLAH-based star retrieval, which is inspired from [Uchiyama and

Saito 2011].

First of all, stars on a captured image are extracted from a background

by thresholding pixel intensity and detecting blobs with labeling.

Correspondences between extracted stars and ones in the

database are matched by star retrieval with LLAH. For fast matching,

dimensional reduction to a 1D index with a hash function is

performed. With matched stars, different constellations on the map

are identified by retrieving each constellation ID. The retrieval process

is performed every time new constellations are detected in

a current frame, and it well supports continuous augmentation of

multiple constellations. Then, a camera pose is estimated using a

homography between correspondences and refined by RANSAC.

With the computed camera pose and the ID, finally, constellation

information is rendered on the map. Note that, stars of each constellation

and their neighbor stars are predefined as references, and

Corresponding author

Figure 1: Visualizing star patterns and mythological illustrations

of retrieved constellations (Orion and Ta***s) on a celestial map.

their image coordinates are stored in a database for LLAH.

In the demonstrations, we designed a celestial map using the

Google Earth software1 and used mythological illustrations offered

from the Hubble Source2. The map had two major constellations

(Orion and Ta***s), and each reference had about 80 stars (constellation

stars with their neighbors). As shown in Figure 1, constellation

information was correctly overlaid on the map by utilizing

only star features. Even in geometric changes and partial occlusions

that normally happened in user interaction, the stars were

robustly tracked due to our keypoint-based approach. The processing

time was average 32 milliseconds when multiple constellations

were identified and tracked simultaneously.

3 Applications and Future Works

In the demonstration, our application was designed as an educational

material like picture book, but it would be promisingly extended

to other applications for constellation visualization in largescale

environments such as observatories or exhibitions. Because a

display type is not also limited to common PC monitors, projectors

or mobile phones could be combined with our approach.

References

NAKAI, T., KISE, K., AND IWAMURA, M. 2006. Use of affine invariants

in locally likely arrangement hashing for camera-based

document image retrieval. In Proc. DAS (LNCS3872), 541–552.

UCHIYAMA, H., AND SAITO, H. 2011. Random dot markers. In

Proc. IEEE VR, 35–38.