Multimedia Data Management

Overview

The work in this area focuses on video databases. We investigate issues related to the modeling, representation and efficient retrieval of video objects. Our video data model is based on video segmentation and salient objects. Video segmentation-based models consider a video database consisting of three types of video sequences: videos, scenes, and shots. The content of each video sequence is captured through selected key frames, which are specializations of images. Salient object-based models extract from video objects of interest along with their audio-visual features and spatio-temporal relationships among them. The combination of the two approaches in one video model allows the capture of a rich set of video characteristics. Within the context of this model, we are investigating indexing structures that can capture the rich video content. The major current research focus is similarity-based trajectory retrieval in video databases. Basically, two issues need to be addressed:

A trajectory extracted from a video can be represented as a sequence of points, which represent absolute or relative positions of a moving object within the video frame. If each video frame is treated as a distinct timestamp, the trajectory can be converted into a multi-dimensional time series data. Existing research on similarity-based retrieval of time series data has focused on one dimensional data such as stock data, temperature data, etc, and can not be directly applied to trajectory data which is multi-dimensional and have particular characteristics.

Researchers

Publications

[University of Waterloo]
University of Waterloo
[Department of Computer Science]
Computer Science
[M. Tamer Özsu's home page]
M. T. Özsu

Copyright © M. Tamer Özsu. All rights reserved.
Last update: January 15, 2006