Output of the computer vision algorithm I designed to detect badges in drone video. The blue and green squares are the top candidate detections for each badge size. The yellow square is the final detection. The grey square is the attention window in which the badge will be searched for in the next video frame. Note the variety of user distance and orientation relative to the drone, as well as varying backgrounds and lighting conditions. The most difficult challenge was to make the detector robust to all of this. The detector runs with a speed of 20 frames per second. Tracking Experiments