The University of Arizona: Maintaining Identity in Automated Tracking of Multiple Animals in Video

High throughput study of animal behavior in lab settings depends on software that automatically tracks multiple moving and interacting animals in video.
Modern multi-animal tracking tools often apply deep learning methods to accurately identify animals and their body parts within frames, but typically struggle to maintain the identitiy of an individual over the duration of a recorded session. Here, I describe a new method that applies a Markovian model of movement with skeletal constraints to neural network-derived per-frame label predictions. The method is implemented in our tool DIPLOMAT*, and produces accurate tracking results with reliable individual continuity.

Series: TRIPODS Seminar
Onlline
Presenter: Travis wheeler, University of Montana

ID: 86322334466
  • Audience: Adult
  • Genre: Mathematics
  • Type: Online, Presentation

The event is finished.

Date

Nov 15 2021
Expired!

Time

12:00 pm - 1:00 pm

Cost

Free

More Info

More Information

Location

Online

Organizer

The University of Arizona College of Mathematics
Phone
(520) 621-6866
Website
https://crr.math.arizona.edu/
More Information
QR Code