A New System for Event Detection from Video Surveillance Sequences - Advanced Concepts for Intelligent Vision Systems - page 110

Fig. 1. REGIMVid platform Architecture
[5], Text-based video retrieval [10,6], event detection [16] and Semantic Access
to Multimedia Data [12]) it was decided to develop a common platform for de-
scriptor extraction that could be used throughout the project. In this paper,
we describe our subsystem for video surveillance indexing and retrieval. The re-
mainder of the paper is organised as follows: a general overview of the toolbox
is provided in Section 2, include a description of the architecture. In section 3
we present our approach to detect and extract of moving objects from video
surveillance dataset. It includes a presentation of different concepts taken care
by our system.We present the combining single SVM classifier for learning video
events in section 4. The descriptors of the visual feature extraction will be pre-
sented in section 5. Finally, we present our experimental results for both event
and concept detection future plans for both the extension of the toolbox and its
use in different scenarios.
2 Our System Overview
In this section, we present an overview of the structure of the toolbox. The system
currently supports extraction of 10 low-level (see section 5) visual descriptors.
The design is based on the architecture of the MPEG-7 eXperimentation Model
(XM), the ocial reference software of the ISO/IEC MPEG-7 standard.
The main objectif of our system is to provide automatic content analysis using
concept/event-based and low-level features. The system (figure 2) first detect and
segment the moving object from video surveillance dataset. In the second step,
it extracts three class of features from the each frame, from a static background
and the segmented objects(the first class from Ω in , the second from Ω out and