<![CDATA[Communication is a fundamental human need, yet individuals with hearing impairments continue to face barriers due to limited access to sign language translation technologies. In Indonesia, the adoption of such technologies remains low, particularly in regions such as Sorong, Southwest Papua, creating a communication gap between the Deaf community and the general public. This study develops a web-based detection system for 36 classes of the Indonesian Sign System (SIBI) using the YOLOv5 algorithm. The dataset consists of 5,682 images of SIBI hand poses with variations in lighting and background, divided into 4,970 training images (87%), 376 validation images (7%), and 335 test images (6%). All data were processed through labeling, preprocessing, augmentation, balancing, and model training. The training was conducted for 150 epochs, and the evaluation results show that YOLOv5 is capable of detecting SIBI signs with significant accuracy. Performance evaluation using a confusion matrix achieved a detection accuracy of 95%, supported by stable precision and recall values and real-time inference performance on common web browsers. Usability testing with 20 respondents indicated satisfaction levels above 72.8%, demonstrating that the system is practical and easy to use. This research presents a validated real-time, web-based SIBI detection system that supports inclusive computer vision applications and enhances accessibility in public services such as education, healthcare, and administrative environments.]]>
Copyrights © 2026
