Beyond the Lab: Accessible AI-Tools for Tracking and Analysing Animal Behaviour in Natural Environments

Artificial intelligence is revolutionising how we interpret and interact with images and videos. From facial recognition systems unlocking phones to programs that analyse medical scans and detect cancerous tissue - AI-driven computer vision has transformed routine tasks in both personal and professional contexts. These advances have made it possible to extract meaning from visual data that would be challenging or impossible to capture with traditional methods. Yet, this progress relies heavily on vast amounts of hand-labelled data to train the underlying models - data that is often unavailable in highly specialised fields.

This talk will provide a general introduction to the use of deep neural networks in image recognition settings and focus on how to overcome data scarcity, using animal behavioural research as a prominent example. Studying animal interactions in natural environments requires highly variable training examples, which are difficult to capture in a controlled lab. Standard computer vision tools often struggle to generalise across the diverse species and unpredictable settings encountered in the wild. To address these issues, the Evolutionary Biomechanics Group employs synthetic data generation based on high resolution 3D scans to train deep neural networks, capable of detecting animals in video recordings and extracting their pose in 3D space. Coupled with advances in real-time tracking, these methods enable the analysis of movements of individuals or large groups of animals to inform the design of walking robots and investigate the impact of insect colonies on their ecosystems.

About the Speaker:

Dr Fabian Plum is a Postdoctoral Researcher at Imperial College London. His focus lies in computer vision and deep learning for large-scale animal behavioural research, particularly in areas where suitable training data is traditionally scarce. As a roboticist within the Department of Bioengineering he's working on building accessible open-source hardware and software tools for ecologists around the globe.