V.N. Alexandrov et al. (Eds.): ICCS 2006, Part III, LNCS 3993, pp. 232 – 239, 2006.
© Springer-Verlag Berlin Heidelberg 2006
A Novel Approach for Similarity Measure Schemes Based
on Multiple Moving Objects in Video Databases
Choon-Bo Shim
1
, Chang-Sun Shin
1
, DongGook Park
1
, and Won-Ho So
2
1
School of Information & Communication Engineering
2
Dept. of Computer Education, Sunchon National University, Sunchon, Jeonnam
540-742, South Korea
{cbsim, csshin, dgpark, whso}@sunchon.ac.kr
Abstract. The general aim of this paper is to study the spatio-temporal
modeling techniques which can efficiently represent multiple moving objects' in
video databases. The traditional schemes only consider direction property, time
interval property, and spatial relationship property for modeling moving objects'
trajectories. But, our scheme also takes into account on distance property,
conceptual location information, and related object information so that we may
improve a retrieval accuracy to measure a similarity between two moving
objects as well as them. As its application, we implement the Content- and
Semantic-based Soccer Video Retrieval (CS
2
VR) system by using MS Visual
C++ and DirectX for indexing and searching on soccer video data. The CS
2
VR
helps users to easily extract the trajectory information of soccer ball form
soccer video data semi-automatically as well as to conveniently retrieve the
results acquired by sketching query trajectory with mouse button.
1 Introduction
The initial research issues on the content-based video retrieval have highly
concentrated on data representation schemes which can efficiently model content
itself extracted from video data [1-5]. However, for handling a large amount of
multimedia data, it is required to provide schemes with good retrieval performance on
a variety of user queries. Thus, we propose a new spatio-temporal modeling technique
which can efficiently represent multiple moving objects in video databases. The
traditional schemes only consider direction property, time interval property, and
spatial relationship property for modeling moving objects' trajectories. However, our
scheme also takes into account on distance property, conceptual location information,
and related object information (e.g. player name having a soccer ball) so that we may
improve retrieval accuracy to measure a similarity between two moving objects as
well as them. Therefore, the proposed spatio-temporal scheme can support content-
based retrieval using moving objects' trajectories as well as semantics-based retrieval
using concepts which are acquired through the co nceptual location information of
moving objects. As its application, we design and implement the Content- and
Semantic-based Soccer Video Retrieval (CS
2
VR) system. Finally, in our performance
study, our scheme yields substantially better retrieval performance compared to
existing related work in term of retrieval effectiveness.