Institutional Repository of Vilnius University
Paukščių ant pastatų aptikimas ir atbaidymas naudojant konvoliucinius neuroninius tinklus ir lazerius /
Abstract
dc:descriptionBirds pose significant threats as disease vectors and instigators of substantial damage to urban buildings and infrastructure. While numerous bird deterrence methods exist, the most prevalent in urban areas - bird spikes - are problematic due to their potential to harm birds and their maintenance difficulty. This paper proposes a novel, humane, and effective bird deterrent: a laser-based system augmented with convolutional neural networks (CNN). Initially, the paper assesses various bird repelling methods and underscores the effectiveness of lasers. Subsequently, it evaluates four CNN architectures to identify one that’s optimally compatible with Edge computing. A dataset consisting of 665 images is collected for bird detection, although the study acknowledges the need for a larger dataset to enhance the system’s safety and performance. Following the analysis, the YOLO architecture was chosen due to its compatibility with the task. Several models, including those from the existing literature and using the collected dataset, were compared. Ultimately, the YOLOv8-m model was selected for its superior speed, accuracy, and compact size. The bird detection and deterrence prototype, as described in this paper, employs a camera and a laser. The camera captures an image, and if the model detects a bird, the laser is activated. During the prototype’s development, it was found that incorporating two cameras or a LiDAR system would substantially enhance the system’s convenience and efficiency. This additional functionality would enable remote setting of the distance between the system and the plane where birds are to be deterred.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Savickas, Tautvilis,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD192847283&prefLang=en_US