Drone Detection with YOLO11
Computer-vision pipeline detecting drones in RGB/IR video on the Anti-UAV-RGBT dataset, fine-tuning YOLO11s end-to-end.
- Dataset
- Anti-UAV-RGBT (RGB + Infrared)
- Base model
- YOLO11s
A three-stage computer-vision pipeline for detecting drones in video.
1. Dataset preparation — extract frames from Anti-UAV-RGBT sequences (paired RGB + infrared, with attributes like fast motion, scale variation, thermal crossover, occlusion and low illumination) and convert bounding-box annotations to normalized YOLO format, including hard-negative frames.
2. Model training — fine-tune yolo11s on the prepared dataset (img size 640, early stopping) with memory-safe settings for single-GPU training.
3. Inference — run the trained weights on a video and write annotated output, tuned for small, low-resolution targets.
Highlights real-world Edge-AI work: detection of small, fast objects under challenging conditions — directly aligned with safety-critical product workflows.