Field Evaluation of a YOLOv8-Based Drone Video Prototype for Real-Time Tiger Detection and Early Warning

Abstract

Human–tiger conflict in plantation landscapes remains a critical safety and conservation issue because encounters between workers and tigers can endanger humans while increasing pressure on endangered tiger populations. This study aims to design and conduct a baseline field evaluation of a YOLOv8-based drone video prototype for real-time tiger detection and early warning. The prototype integrates drone-based RGB video acquisition, wireless video transmission, edge-based visual inference, detection logging, and warning output into a single prototype workflow. This study used a systems engineering approach and applied experimentation. The YOLOv8 model was trained using annotated tiger image data and then integrated into the prototype. Field testing was conducted in an open-field baseline scenario using six tiger replicas under two lighting conditions, daytime and evening, to support safety, ethical control, and experimental consistency. System performance was evaluated using precision, recall, F1-score, false negatives, detection range, confidence score, time-to-first-alert, and bounding box stability. The results show that the prototype performed better during daytime testing, achieving 96.90% precision, 85.62% recall, a 90.91% F1-score, a 35 m maximum detection range, 0.60–0.75 average confidence, and a time-to-first-alert of less than 1 s. In evening testing, performance decreased to 93.57% precision, 55.36% recall, 69.57% F1-score, a 7 m maximum range, 0.40–0.55 average confidence, and 1.8–2.5 s response time. These findings indicate that the prototype provides an initial technical basis for drone-based early warning, but further validation is required using real tiger data, complex plantation environments, higher occlusion levels, and improved low-light sensing before operational deployment can be claimed.

GS Cited Analysis