Összes szerző


Berekméri Evelin

az alábbi absztraktok szerzői között szerepel:

Berekméri Evelin
Investigation of a red-footed falcon hub in Angola with deep learning

Aug 29 - kedd

15:30 – 17:00

I. Poszterszekció

P04

Investigation of a red-footed falcon hub in Angola with deep learning

Evelin Berekméri1,2, Nico Klar,3,4, Eric Price4,5, Péter Palatitz6, Aamir Ahmad4,5, Máté Nagy1,2,7

1Department of Biological Physics, Eötvös Loránd University, Budapest, Hungary

2MTA-ELTE Lendület Collective Behaviour Research Group, Hungarian Academy of Sciences, Budapest, Hungary

3Center for Solar Energy and Hydrogen Research Baden-Württemberg, Germany

4Institute of Flight Mechanics and Controls, University of Stuttgart, Stuttgart, Germany

5Max Planck Institute for Intelligent Systems, Tübingen, Germany

6MME/BirdLife Hungary, Budapest, Hungary

7Max-Planck Institute of Animal Behavior, Konstanz, Germany

Automating the perception of our environment with object detection has an increasing number of applications, not only in everyday life, such as for self-driving cars or in security, but also in scientific research. Object detection is a deep learning-based computer vision technique that involves recognising predefined classes of objects in visual recordings and locating them within the frame of the recording.

In this study, we focus on a recently discovered, yet unpublished migratory bird hub in Angola, where red-footed falcons gather from different points of the world. These hubs play a crucial role in connecting remote locations worldwide, potentially assisting in disease transmission, and acting as indicators of global environmental changes due to the birds' sensitivity to environmental variations.

Our aim is to identify and track birds on video recordings from the hub providing insight into their population size and preparing further collective behaviour studies such as identifying when they are making use of thermals to gain altitude during their flights. We utilise state-of-the-art object detection frameworks and utilize the latest developments in deep learning and computer vision, taking into consideration that we also aim to deploy our method for real-time, on-board detection on drones.

Currently our understanding of the population size of this species, based on traditional counting methods, has multiple orders of magnitude uncertainty. Our approach offers numerous advantages over traditional procedures, being not only faster and more cost-efficient but also can be more accurate and it also offers real-time monitoring. Such approaches have been used to study the population of other species, such as bat colonies.

Acknowledgment

PREPARED WITH THE PROFESSIONAL SUPPORT OF THE DOCTORAL STUDENT SCHOLARSHIP PROGRAM OF THE CO-OPERATIVE DOCTORAL PROGRAM OF THE MINISTRY OF INNOVATION AND TECHNOLOGY FINANCED FROM THE NATIONAL RESEARCH, DEVELOPMENT AND INNOVATION FUND.

E.B. acknowledges MTA-ELTE Lendület Collective Behaviour Research Group for financial suppport.

This project was partially supported by the National Research, Development and Innovation Office under grant no. K128780.

Nagy Máté
High-throughput collective animal behavior studies and their connections to artificial systems

Aug 29 - kedd

11:30 – 11:45

Modern biofizikai módszerek

E10

High-throughput collective animal behavior studies and their connections to artificial systems

Máté Nagy1,2,3, Evelin Berekméri1,2, Pedro Lacerda1,2, Göksel Keskin1,2, Zoltán Szarvas1,2, Péter Palatitz1, Tamás Nepusz2, Gábor Vásárhelyi2

1 MTA-ELTE Lendület Collective Behaviour Research Group, Hungarian Academy of Sciences, Budapest, Hungary
2 Department of Biological Physics, Eötvös Loránd University, Budapest, Hungary

3 Max-Planck Institute of Animal Behavior, Konstanz, Germany

Spectacular aerial displays of birds, the mesmerizing swirling of giant schools of fish or the rumbling gallop of hundreds of horses are fascinating examples of group behavior occurring in nature. During the presentation, we examine how these interesting phenomena are created through the interactions of individuals with each other and their environment, and how we can understand them with the help of physics and emerging technologies. What ways can we measure and quantify behavior using high-throughput methods and deep learning. Virtual reality for animals provides a tool to study decision making, and to test hypotheses through animals interacting with computational models. Can robots using artificial intelligence achieve similar (or even better) performance as animals? Let's see what we can learn from the biological insights and how we can use them to design improved artificial systems.