Wild Animal Detection using YOLOv8
Brahm Dave, Meet Mori, Anurag Bathani, Parth Goel
Abstract
People residing on the outskirts of forests encounter numerous challenges in their daily lives due to conflicts between humans and wild animals. This conflict has resulted in the loss of lives among both the local population and farmers, as well as substantial livestock casualties. The Gujarat Forest Department has reported that 12 individuals have lost their lives, 70 have been injured, and 3927 cattle have been killed. To address this conflict, various government initiatives have been implemented to track and monitor wild animals. This paper presents a deep learning-based model to track wild animals in real-time from camera footage. In this study, we propose the utilization of the YOLOv8 architecture to detect four distinct categories: Lions, Tigers, Leopards, and Bears. The dataset is constructed from various documentaries, YouTube videos, and existing datasets from Kaggle. This dataset contains 1619 images with annotated four categories of objects. We trained three different YOLOv8 architectures (medium - YOLOv8m, large - YOLOv8l, and extra-large - YOLOv8x) on our dataset. To improve the accuracy of our model, we applied various types of augmentation to the dataset images. The trained extra-large model achieved a mAP of 94.3%, demonstrating its effectiveness in detecting wild animals in real time at 20 FPS.