Hand Gesture Recognition for Filipino Sign Language Under Different Backgrounds

Abstract

The article implements a hand gesture Filipino Sign Language recognition model using Raspberry Pi. Numerous studies on Filipino Sign Language (FSL) frequently identify a letter with a glove and using a plain background, which may be challenging if implemented in a more complex background. Limited research on the implementation of YOLO-Lite and MobileNetV2 on FSL were also observed. Implementing YOLO-Lite for hand detection and MobileNetV2 for classification, the average accuracy achieved for differentiating 26 hand gestures, representing FSL letters, was 93.29%. The model demonstrated dependability in a variety of complex backgrounds. However, challenges in recognizing letters Q, J, and Z were encountered. Additionally, in letters N and M, due to their similar hand structures, N is sometimes mistakenly interpreted as M. The model developed by the researchers performed well and demonstrated better accuracy compared to a different model. The system was able to achieve higher accuracy while running on limited resources and in various environments.