A new paper on detecting and tracking elephants in drone videos
A framework for detecting and tracking elephants in drone videos
By Chaim Chai Elchik, Serge Wich, and André Burger
The escalating global biodiversity crisis requires innovative, scalable solutions to monitor wildlife populations. In a new open access paper in Drone Systems and Applications, we introduce a framework that utilizes drone video streams to automate the detection and tracking of African elephants (Loxodonta africana).
Our approach integrates state-of-the-art object detection (YOLOv11) and tracking (BoT-SORT) methods. Crucially, we enhanced these standard tools with a custom post-track re-identification algorithm. This novel step is designed to mitigate “identity switching,” a common issue where trackers lose an animal’s identity during occlusions or rapid camera movements.
Tested on footage from the Welgevonden Game Reserve in South Africa, the framework significantly improved tracking consistency, raising the association accuracy (AssA) score to 0.912. Beyond simple counting, the system generates key behavioral metrics, including individual movement speeds, trajectories, and herd cluster statistics.
By automating data processing, this tool offers ecologists a way to extract deep insights from drone footage with significantly reduced manual labor.
The code is hosted on our brand new GitHub page: https://github.com/ConservationDronesAI/ElephantDetectionAndTracking