Extraction of recurring behavioral motifs from video recordings of natural behavior
Martin Ståhl · KTH Publication Database DiVA (KTH Royal Institute of Technology) · 2018
Complex neural activity exhibits itself in various forms, one of which is behavior. Hence a natural way to study neural activity is to analyze behavior. In this thesis, behavior has been studied using a Gaussian hidden Markov model. The data has been gathered from video recordings of free roaming mice in a box. The model has trained on and classified mouse behavior. Classification with 4 and 6 states have been tried, the one with 6 states seems to make a distinction between two different stationary states which is biologically interesting. The conclusion is that the Gaussian hidden Markov model is a reasonable approach to mice behavior classification but it does not solve any fundamental problems. There are also some data gathering techniques that affect the results which need to be improved.