idTracker: tracking individuals in a group by automatic identification of unmarked animals
Résumé
Animals in groups touch each other, cross or interact in complex ways. Current videotracking methods sometimes switch identities of unmarked individuals during these interactions. These errors propagate and result in random assignments after few minutes unless manually corrected. We present idTracker, a multitracking algorithm that from a video recording of a group extracts a fingerprint that characterizes each animal. It then uses these fingerprints to identify every individual throughout the same video. Tracking by identification prevents propagation of errors, correct identities being kept for an indefinite time. idTracker distinguishes animals even when humans cannot, as for size-matched siblings, and reidentifies animals after they temporarily disappear from view, or in different videos. It is robust, easy to use and general. We tested it on fish (Danio rerio and Oryzias latipes), flies (Drosophila melanogaster), ants (Messor structor) and mice (Mus musculus).
Origine | Fichiers produits par l'(les) auteur(s) |
---|