Speaker
Description
Studying freely moving animals, as opposed to restrained or head-fixed animals, provides an excellent opportunity to study a wider range of complex, natural behaviours and interactions in neuroscience beyond the confines of the traditional laboratory setting. Advances in motion tracking technology, with the help of machine learning, are beginning to make the study of natural behaviours possible without the need for laborious hand-scoring or restrictive head positions. Here, we describe a motion-tracking setup for a marmoset home cage, which records daily behaviour and interactions. An array of sixteen cameras record a group of marmosets moving freely in their home cage, without physical markers. Recording several hours of video per day creates a rich but very large dataset, which would require a tremendous amount of labour to analyse by hand. We are therefore utilising emerging machine-learning techniques to extract pose and behaviour information. From video footage, we will first extract three-dimensional pose information from multiple animals, as well as identity information, using machine learning approaches. We will further use machine learning to identify instances of defined behaviours without the need to label entire datasets manually. This approach, combined with wireless neural recorders, will facilitate large-scale analysis of complex and natural behaviours and their neuronal basis with less restrictive experimental setups.
