University of Twente Student Theses
Automated Detection of Provisioning Events and Prey in Nestbox Video Footage
Visser, Jesse Wiebe (2022) Automated Detection of Provisioning Events and Prey in Nestbox Video Footage.
|
PDF
1MB |
| Abstract: | To monitor the provisioning behaviour of cavity breeding bird species, camera enhanced nestboxes are used to capture video footage of the birds entering and exiting their nestboxes. These setups often collect enormous amount of data, which researchers are tasked to go through to find the provisioning events and categorize the prey being provisioned. This process is time consuming and prone to human error. Instead using Convolutional Neural Networks (CNN) this process can be largely automated. For this task two CNN models were trained on a dataset of more than 48 hours of nestbox video footage of Redstarts (Phoenicurus phoenicurus) and their young. The 80.000 of the video frames were annotated to be either provisioning or not-provisioning and a subset of 1270 frames was annotated with the location and category of the prey provisioned by the parent. The first MobileNetV3Small model classifies frames as either provisioning or not-provisioning with an accuracy of 98.75% at 290fps and the second YOLOR model detects the general location of the prey and classifies it into one of four categories, Caterpillar, Moth, Spider or Other. The model was able to run inference at 90fps with a mAP of 83.6% at an IoU threshold equal to 0.5. |
| Item Type: | Essay (Bachelor) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 43 environmental science, 54 computer science |
| Programme: | Creative Technology BSc (50447) |
| Link to this item: | https://purl.utwente.nl/essays/91875 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page