ELTE biophysicists have developed a fast, accurate, and field-deployable algorithm for identifying individual animals. The RAPID system—created in collaboration between the MTA–ELTE “Lendület” Collective Behaviour Research Group, the Max Planck Institute, and the University of Stuttgart—is capable of recognizing individual patterned animals in real time, even from drone footage. The new method represents a significant advance in the study of individual behavior and in wildlife monitoring.
Identifying individual animals is a key challenge in ecology and behavioral research. For a long time, field observations relied primarily on the manual work of experts. However, with the development of artificial intelligence and computer vision, it has become possible to automate this task. While most existing solutions achieve high accuracy, they are often not suitable for real-time use or for deployment on devices with limited computational capacity in field conditions. The RAPID (Real-time Animal Pattern re-Identification on Edge Devices) algorithm addresses this challenge. It was developed by András Zábó and Máté Nagy of the MTA–ELTE “Lendület” Collective Behaviour Research Group, together with Aamir Ahmad of the Max Planck Institute and the University of Stuttgart.
A key advantage of RAPID is its ability to operate in real time: it can process 40–60 identifications per second on a standard computer, and up to 10 identifications per second on smaller, embedded devices such as those used in drones. This means the system is suitable not only for post-processing but also for live monitoring, for example as part of autonomous drone-based observation systems.
The algorithm relies on the unique visual patterns of animals. The stripes of zebras, the spots of giraffes, or the markings of jaguars differ from one individual to another, much like human fingerprints. RAPID analyzes these visual features and compares them to a previously built image database, enabling the identification of specific individuals.
The efficiency of the system is further enhanced by the fact that it does not require high-performance graphical processing units: it operates solely on CPU while maintaining accuracy comparable to state-of-the-art methods. Another important feature of the development is that each identification is accompanied by a confidence score, which supports researchers in interpreting the results.
RAPID has been successfully validated on multiple datasets and across a range of species, including zebras, giraffes, tigers, and other patterned animals. It also demonstrated reliable performance under diverse environmental and technical conditions, such as varying lighting, camera systems, and acquisition settings.
The new method has the potential not only to advance research but also to expand the practical toolkit of wildlife conservation. Combined with drone technology and artificial intelligence, it enables continuous, non-invasive monitoring of wild animal populations, contributing to more accurate population assessments and more effective protection of endangered species.
RAPID is available as an open-source development, allowing the international research community to access and further improve the system. Ongoing work focuses on extending its capabilities, including the identification of individuals not yet present in the database and handling variations in viewing angles.
____________________________________________________
The publication is available here.
Further information is available on the University of Stuttgart’s website.
The cover image is for illustration purposes (pexels.com).