학술논문

Enumeration of Birds using Video Segmentation for a Better Understanding of Bird Behaviors
Document Type
Conference
Source
2023 IEEE Integrated STEM Education Conference (ISEC) Integrated STEM Education Conference (ISEC), 2023 IEEE. :179-186 Mar, 2023
Subject
Bioengineering
Communication, Networking and Broadcast Technologies
Engineering Profession
General Topics for Engineers
Robotics and Control Systems
Signal Processing and Analysis
YOLO
Training
Tracking
Birds
Behavioral sciences
Statistics
Videos
Video Segmentation
HOG
SSD
R-CNN
Bird Behavior
Climate Change
Sustainability
Ecology
Language
ISSN
2473-7623
Abstract
The State of the World’s Birds 2022 report estimated a decline in 30% of the world’s birds since 1970, largely attributed to habitat loss, climate change, and other anthropogenic factors [1, 3]. Chimney swifts are aerial insectivores that forage during the day and commonly roost in large colonies inside hollow vertical human-made structures such as chimneys. Signaled by sunset, large groups of this species, sometimes comprised of thousands of individuals, exhibit an impressive display of coordinated entry into their roosting sites. Despite this striking behavior and close relationship with humans, the factors influencing this collective behavior are poorly understood. To understand how anthropogenic and environmental factors influence the population health of this (and other species), reliable methods for counting individuals are required. The purpose of this study is to develop an automated method to accurately count the number of birds over a given period at the roosting site.To understand the behavior of colonially roosting species such as the Chimney swift, we proposed a methodology in which we collect and analyze videos of birds at roosting sites to understand their behaviors. The proposed approach consists of two steps: physically gathering the video with cameras perched at optimal locations [2] and utilizing video segmentation algorithms to count the number of birds at roosting sites over time.The final goal of this research is to build useful data for ecologists that can be utilized to fill critical gaps in knowledge. To achieve this, the number of birds counted over time will be analyzed with ambient weather datasets at different geographical locations. Therefore, in this paper, we use four video segmentation algorithms to detect and track the movement of birds, count the number of birds at certain intervals, and then compare the performance of the algorithms.