Globally Optimal Object Tracking with Complementary Use of Single Shot Multibox Detector and Fully Convolutional Network

研究成果: Chapter in Book/Report/Conference proceedingConference contribution

抄録

Object tracking is one of the most important but still difficult tasks in computer vision and pattern recognition. The main difficulties in the tracking task are appearance variation of target objects and occlusion. To deal with those difficulties, we propose a object tracking method combining Single Shot Multibox Detector (SSD), Fully Convolutional Network (FCN) and Dynamic Programming (DP). SSD and FCN provide a probability value of the target object which allows for appearance variation within each category. DP provides a globally optimal tracking path even with severe occlusions. Through several experiments, we confirmed that their combination realized a robust object tracking method. Also, in contrast to traditional trackers, initial position and a template of the target do not need to be specified. We show that the proposed method has a higher performance than the traditional trackers in tracking various single objects through video frames.

本文言語英語
ホスト出版物のタイトルImage and Video Technology - 8th Pacific-Rim Symposium, PSIVT 2017, Revised Selected Papers
編集者Carlos Hitoshi, Manoranjan Paul, Qingming Huang
出版社Springer Verlag
ページ110-122
ページ数13
ISBN(印刷版)9783319757858
DOI
出版ステータス出版済み - 2018
イベント8th Pacific-Rim Symposium on Image and Video Technology, PSIVT 2017 - Wuhan, 中国
継続期間: 11 20 201711 24 2017

出版物シリーズ

名前Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
10749 LNCS
ISSN(印刷版)0302-9743
ISSN(電子版)1611-3349

その他

その他8th Pacific-Rim Symposium on Image and Video Technology, PSIVT 2017
国/地域中国
CityWuhan
Period11/20/1711/24/17

All Science Journal Classification (ASJC) codes

  • 理論的コンピュータサイエンス
  • コンピュータ サイエンス(全般)

フィンガープリント

「Globally Optimal Object Tracking with Complementary Use of Single Shot Multibox Detector and Fully Convolutional Network」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル