Drone Detection Annotation — Visual and Thermal Video
Challenge
The client was building AI for detecting drones and aircraft in two distinct video types: standard visual footage and thermal infrared feeds. Thermal video presents unique annotation challenges — objects appear as heat signatures rather than recognizable shapes, requiring annotators to learn an entirely different visual language.
Solution
We processed two separate batches: 64 visual-spectrum drone videos and 35 minutes of thermal footage. Annotators placed bounding boxes on aerial objects across both modalities, with quality protocols adapted for the distinct characteristics of each video type.
Results
Delivered 129 hours of annotation across both video types. The quality of thermal video annotation — a particularly specialized skill — led to continued collaboration on subsequent defense-grade datasets.