跳到主要导航 跳到搜索 跳到主要内容

Key object discovery and tracking based on context-aware saliency

  • Xi'an Jiaotong University

科研成果: 期刊稿件 › 文章 › 同行评审

10 引用 (Scopus)

摘要

In this paper, we propose an online key object discovery and tracking system based on visual saliency. We formulate the problem as a temporally consistent binary labelling task on a conditional random field and solve it by using a particle filter. We also propose a context-aware saliency measurement, which can be used to improve the accuracy of any static or dynamic saliency maps. Our refined saliency maps provide clearer indications as to where the key object lies. Based on good saliency cues, we can further segment the key object inside the resulting bounding box, considering the spatial and temporal context. We tested our system extensively on different video clips. The results show that our method has significantly improved the saliency maps and tracks the key object accurately.

源语言英语
期刊论文编号18
期刊International Journal of Advanced Robotic Systems
卷10
DOI
出版状态已出版 - 8 1月 2013

学术指纹

探究 'Key object discovery and tracking based on context-aware saliency' 的科研主题。它们共同构成独一无二的学术指纹。

引用此