Deep Learning based Animal Detection and Tracking in Drone Video Footage
Source of Publication
ACM International Conference Proceeding Series
In this paper, we propose a multiple animal tracking system in drone footage that is designed and implemented using a Deep Neural Network (DNN) based tracking-by-detection approach. The proposed system consists of two main components, namely the sub-system for animal detection, and the sub-system for animal tracking. In the animal detection component, we exploit the effective use of YOLO-V5 to detect individual animals and in the tracking component, we use a centroid tracking algorithm to associate the location of the detected animals in consecutive video frames. The performance of the proposed system is analyzed on drone video footage containing herds of Arabian Oryx with complex patterns of movement of individual animals. All videos were recorded by using a drone flying over known oryx feeding points in the desert areas of the UAE. The experimental results showed that our tracking system can detect and track individual oryxes within herds, accurately, even when the oryxes are very close to each other, partially occluded and their walking paths cross each other.
animal tracking, centroid tracking algorithm, drone video analytics, Tracking-by-detection, YOLO-V5
Jintasuttisak, Thani; Leonce, Andrew; Sher Shah, Moayyed; Khafaga, Tamer; Simkins, Greg; and Edirisinghe, Eran, "Deep Learning based Animal Detection and Tracking in Drone Video Footage" (2022). All Works. 5253.
Indexed in Scopus