نوع مقاله : مقاله مستقل
عنوان مقاله English
نویسندگان English
Automatic conversion of sign language into text plays an important role in facilitating communication between deaf individuals and society. In this research, a real-time system for recognizing static American Sign Language letters is presented. To create the dataset, more than 68,000 samples of 24 static American Sign Language letters were collected by two individuals using a webcam and MediaPipe under various conditions including changes in hand angle, scale, and orientation. In the feature extraction stage, 21 hand key points along with geometric features including angles between fingers and distances from joints to the wrist were extracted. The dataset was divided into training and test sets with an 80/20 split, followed by standardization. In the classification stage, a Support Vector Machine with Bayesian optimization was used, achieving an accuracy of 99.05% on the test set drawn from the same two individuals. In an independent evaluation on a third individual whose samples were never used in training or tuning, the accuracy dropped to 82.27%, indicating limited generalization to new users. Furthermore, to improve output readability, a post-processing layer including spell correction and keyword extraction was added. The proposed system provides high accuracy under conditions similar to the training data and readable output; however, it is limited to static letters and its generalization to unseen signers requires a larger and more diverse dataset.
کلیدواژهها English