Publication detail

System for Object Detection and Tracking Using Compressed Domain

HASMANDA, M. BENEŠ, R. ŘÍHA, K.

Original Title

System for Object Detection and Tracking Using Compressed Domain

English Title

System for Object Detection and Tracking Using Compressed Domain

Type

conference paper

Language

en

Original Abstract

This paper presents possibilities of using motion vectors included in encoded MPEG or H.264 videos to detect and track various objects, especially people. Currently, there are techniques based on optical flow applied on segmented images. Such methods often have high time complexity, which can complicate their utilization for real-time applications. In addition, most of segmentation based methods fail when multiple objects "which are to be detected" are overlapping. In the proposed approach, motion vectors that were already calculated during video compression are utilized. This new approach could be useful when processing the data stream directly from a camera - in such a case algorithm speed is an essential criterion. The main disadvantage of this approach resides in its dependency on the accuracy of the compression algorithm of encoder, which calculates motion vectors. The proposed method was tested on a real video-sequences containing moving object captured by common cameras.

English abstract

This paper presents possibilities of using motion vectors included in encoded MPEG or H.264 videos to detect and track various objects, especially people. Currently, there are techniques based on optical flow applied on segmented images. Such methods often have high time complexity, which can complicate their utilization for real-time applications. In addition, most of segmentation based methods fail when multiple objects "which are to be detected" are overlapping. In the proposed approach, motion vectors that were already calculated during video compression are utilized. This new approach could be useful when processing the data stream directly from a camera - in such a case algorithm speed is an essential criterion. The main disadvantage of this approach resides in its dependency on the accuracy of the compression algorithm of encoder, which calculates motion vectors. The proposed method was tested on a real video-sequences containing moving object captured by common cameras.

Keywords

motion vector, object detection, object tracking, MPEG.

RIV year

2012

Released

12.09.2012

Publisher

University of Zilina

Location

Zilina

ISBN

978-80-554-0569-8

Book

The 14th International Conference on Research in Telecommunication Technologies RTT - 2012

Edition

1

Pages from

259

Pages to

263

Pages count

5

Documents

BibTex


@inproceedings{BUT93851,
  author="Martin {Hasmanda} and Radek {Beneš} and Kamil {Říha}",
  title="System for Object Detection and Tracking Using Compressed Domain",
  annote="This paper presents possibilities of using motion vectors included in encoded MPEG or H.264 videos to detect and track various objects, especially people. Currently, there are techniques based on optical flow applied on segmented images. Such methods often have high time complexity, which can complicate their utilization for real-time applications. In addition, most of segmentation based methods fail when multiple objects "which are to be detected" are overlapping. In the proposed approach, motion vectors that were already calculated during video compression are utilized. This new approach could be useful when processing the data stream directly from a camera - in such a case algorithm speed is an essential criterion. The main disadvantage of this approach resides in its dependency on the accuracy of the compression algorithm of encoder, which calculates motion vectors. The proposed method was tested on a real video-sequences containing moving object captured by common cameras.",
  address="University of Zilina",
  booktitle="The 14th International Conference on Research in Telecommunication Technologies RTT - 2012",
  chapter="93851",
  edition="1",
  howpublished="electronic, physical medium",
  institution="University of Zilina",
  year="2012",
  month="september",
  pages="259--263",
  publisher="University of Zilina",
  type="conference paper"
}