Related Experiment Video
Updated: Jul 1, 2026
08:14
Who is Who? Non-invasive Methods to Individually Sex and Mark Altricial Chicks
Published on: May 24, 2014
18.5K
Persistent animal identification leveraging non-visual markers
Michael P J Camilleri1, Li Zhang2, Rasneer S Bains3
1School of Informatics, University of Edinburgh, Edinburgh, UK.
Machine Vision and Applications
|July 17, 2023
Summary
This study introduces a novel method for uniquely identifying individual mice in cluttered environments using RFID and tracking data. The approach achieves 77% accuracy, enabling automated behavior recognition in biological research.
Area of Science:
- Computer Vision
- Machine Learning
- Animal Behavior Analysis
Background:
- Automated behavior recognition in biological research requires accurate individual animal identification.
- Tracking mice in cluttered home-cage environments is challenging due to lack of visual features and frequent occlusions.
Purpose of the Study:
- To develop a robust method for unique mouse identification over time in cluttered environments.
- To enable automated behavior recognition by solving the animal identification problem.
Main Methods:
- Formulated animal identification as an assignment problem solved with Integer Linear Programming.
- Developed a novel probabilistic model integrating visual tracklets with coarse RFID location data.
- Created a curated dataset with ground-truth annotations for model evaluation.
Main Results:
- Achieved 77% accuracy in identifying individual mice.
- Successfully rejected spurious detections when animals were hidden.
- Demonstrated the potential of combining weak tracking with coarse identity information.
Conclusions:
- The proposed method effectively addresses the challenge of individual mouse identification in complex environments.
- This approach is a crucial step towards reliable automated behavior recognition in laboratory animals.
- The developed probabilistic model provides a principled way to handle object detection with coarse localization.
