Tracking Golden-Collared Manakins in the Wild (bibtex)
by Anna Gostler
Abstract:
Male golden-collared manakins are tropical birds that perform an elaborate courtship display which determines their mating success. Biologists recorded the birds’ displays in the jungle with high-speed cameras. To analyze what constitutes a good courtship performance the biologists use the bird’s trajectory, which they currently obtain by manually annotating the videos frame by frame. Automatically tracking the bird can save a lot of time. The videos of the courtship displays are challenging for a tracker: the bird is susceptible to motion blur, quickly changes its appearance and often leaves the frame. The cluttered background contains elements that visually resemble or occlude the bird. We present an online visual tracking algorithm, which combines a Mixture of Gaussians model to detect moving objects, a Convolutional Neural Network trained to recognize the male golden-collared manakin, and a Kalman Filter as a motion model. Our tracker achieves better accuracy and robustness on a dataset of videos of courtship displays than state-of-the-art trackers.
Reference:
Tracking Golden-Collared Manakins in the Wild (Anna Gostler), (Nicole M. Artner, ed.), Technical report, PRIP, TU Wien, 2018.
Bibtex Entry:
@TechReport{TR141,
  author =	 "Anna Gostler",
  editor = "Nicole M. Artner",
  title =	 "Tracking Golden-Collared Manakins in the Wild",
  institution =	 "PRIP, TU Wien",
  number =	 "PRIP-TR-141",
  year =	 "2018",
  url =		 "https://www.prip.tuwien.ac.at/pripfiles/trs/tr141.pdf",
  abstract =	 "Male golden-collared manakins are tropical birds that perform an elaborate courtship display which determines their mating success. Biologists recorded the birds’ displays in the
jungle with high-speed cameras. To analyze what constitutes a good courtship performance
the biologists use the bird’s trajectory, which they currently obtain by manually annotating the videos frame by frame. Automatically tracking the bird can save a lot of time.
The videos of the courtship displays are challenging for a tracker: the bird is susceptible
to motion blur, quickly changes its appearance and often leaves the frame. The cluttered
background contains elements that visually resemble or occlude the bird. We present an
online visual tracking algorithm, which combines a Mixture of Gaussians model to detect
moving objects, a Convolutional Neural Network trained to recognize the male golden-collared manakin, and a Kalman Filter as a motion model. Our tracker achieves better
accuracy and robustness on a dataset of videos of courtship displays than state-of-the-art
trackers.",
}
Powered by bibtexbrowser