Real-time sign language gesture recognition for Women's safety using dynamic time warping and voting algorithms

Advancements in computer vision techniques have spurred the development of sign language recognition systems, benefiting the deaf and mute community. However, the safety of deaf hard-of-hearing women remains a significant concern, particularly in the context of emerging threats. This paper presents a method for real-time recognition of women's safety sign language gestures using Dynamic Time Warping (DTW) and voting algorithms. The proposed approach leverages MediaPipe to extract keypoints from hand, body, and facial movements, representing gestures as feature matrices that encapsulate both spatial and temporal aspects. These matrices are compared using DTW to recognise gestures, with a voting algorithm determining the final gesture label. A novel dataset, WSISL-28, consisting of 28 women's safety-related gestures, was created to evaluate the method. Experimental results demonstrate an impressive accuracy of 98.45 %, affirming the feasibility and effectiveness of the proposed system for real-time sign language gesture recognition, which empowers deaf and mute women, enhancing their safety and security.