Pedestrian Tracking via an On-line Boosting Method

DOI Web Site
  • 王 遷
    Interdisciplinary Graduate School of Science and Engineering, Tokyo Institute of Technology
  • 青木 工太
    Imaging Science and Engineering Laboratory, Tokyo Institute of Technology
  • 長橋 宏
    Imaging Science and Engineering Laboratory, Tokyo Institute of Technology

この論文をさがす

抄録

In actual surveillance conditions, there exist a lot of uncertainties in pedestrian movements. These movements may disturb most tracking algorithms and result in tracking failure. In this paper, a new pedestrian tracking system is proposed, in which an online-boosting method is embedded into the pedestrian tracking of object model. Online-boosting [1] methods have become very popular in computer vision, and achieved impressive performance in detection and recognition tasks. However, the on-line adaption has to face one key problem: an error which finally leads to tracking failure (drifting) may be accumulated in each update of the tracking. In order to avoid the drifting problem, MIL (Multiple Instance Learning) method is proposed, instead of traditional supervised learning, to lead to a more robust tracker with well performance.

収録刊行物

キーワード

詳細情報 詳細情報について

問題の指摘

ページトップへ