Real-time Pet Identification Using Computer Vision on a Raspberry Pi
Name
Villem Susi
Abstract
This work investigates using deep convolutional neural networks to detect and identify domestic pets using camera-captured images on an edge device - Raspberry Pi. To do this, we use a pre-trained YOLOv5 to detect pets and a fine-tuned EfficientNet on a modified dataset for the identification of specific pets. Furthermore, a user interface is created to add additional pets, view logs, fine-tune the model and interact with the device. We describe a custom dataset of internet-sourced and author-provided cat and dog images. This work obtains three F1 scores with three different datasets on the identification task: 0.25, 0.67 and 0.84. The results highlight a need to improve the reliability of the identification process.
Graduation Thesis language
English
Graduation Thesis type
Bachelor - Computer Science
Supervisor(s)
Karl Kaspar Haavel
Defence year
2024