Video-based tracking

Contact person: Anton S. Konushin (ktosh@graphics.cs.msu.ru)

Video-based people tracking

An aim of the project is to create video based people tracking algorithm. We define tracking as a estimation of person location in each frame and for each person presented in the video sequence. Currently the most promising methods of multiple person tracking are based on tracking-by-detection approach. Within this framework algorithms consist of following three stages:

  1. People detection in the key frames of the video sequence
  2. People trajectories construction
  3. Person location estimation in the all frames of the video sequence

On the first stage of the pipeline person head detector can be applied to detect people presented in the frame. Some persons may not be found because of errors of the detector. Also detector may not be applied to each frame through its high complexity:

But as a result it is necessary to estimate people location in each frame of the video sequence:

Team

Publications

Acknowledgements

This work is partillay supported by grant RFBR 14-01-00849 "People detection, tracking and annotation in video streams" (2014-2016), grant RFBR 11-01-00957-а "Video-based person analysis algorithms for smart interfaces" (2011-2013) and grant of the President of the Russian Federation for the young scientists МК-4644.2012.9 "Content-based search in video archives".