<![CDATA[This study proposes a real-time Android-based American Sign Language (ASL) gesture detection system using YOLO26. The model recognizes 26 static gesture classes consisting of 24 alphabet gestures (a-i and k-y) and two common expressions, namely “I love you” and “thank you”, from images captured by a smartphone camera. A custom dataset of 7,800 images was prepared and divided into 5,460 training images, 1,560 validation images, and 780 testing images. Preprocessing included resizing, normalization, horizontal flipping, random rotation, brightness adjustment, contrast variation, and zoom augmentation to improve robustness under different acquisition conditions. Three YOLO26 variants, namely YOLO26s, YOLO26n, and YOLO26m, were trained and evaluated using precision, recall, F1-score, mAP@50, mAP@50-95, latency, frames per second, and lighting robustness. Experimental results show that YOLO26n provided the most balanced deployment performance with 96.4% precision, 95.8% recall, 96.1% F1-score, 97.8% mAP@50, and 84.2% mAP@50-95. Real-time testing on an Infinix X6726 device produced an average latency of 118 ms per frame or 8.47 FPS. Robustness testing under high, medium, and low lighting produced detection success rates of 98.3%, 96.7%, and 93.3%, respectively. The findings indicate that YOLO26 is feasible as an exploratory architecture for Android-based ASL gesture detection, although broader cross-device and cross-dataset validation is still required before claiming general real-world superiority.]]>
Copyrights © 2026
