Video-based Hand Gesture Recognition using Random Forest for Sign Language Interpretation

• Published in: 2024 Asia Pacific Conference on Innovation in Technology (APCIT) pp. 1 - 6
• Main Authors: Himasree, J, P L, Jeevitha, K, Deekshitha, Kolisetty, Aashrita, Naveen, Soumyalatha
• Format: Conference Proceeding
• Language: English
• Published: IEEE 26-07-2024
• Subjects: Accuracy; Codes; Computer vision; human computer interaction; Libraries; Python; Random Forest; Reliability; sequential pattern; Sign language; Technological innovation; Training; Video capturing
• DOI: 10.1109/APCIT62007.2024.10673591

Effective communication is essential for all individuals, yet it can be particularly challenging for the deaf community, as sign language is their major means of communication. We are researching how to create a reliable and effective system that can understand sign language motions correctly without the help of a human, realizing the importance of this problem.In this research work we have chosen to construct our own dataset rather than depending on pre-existing ones that may be accessed online because we understand how important authenticity and relevance are. This method gives us more control over the quality of the data and guarantees that the dataset closely matches our study goals. In addition, our system can be readily expanded and recognizes a wide range of hand motions and signs, including the alphabet. We can support a broad variety of sign language expressions thanks to this versatility, which improves inclusivity and usability. Random Forest is used to encode temporal variations among motions in sign language. The system's overall accuracy and reliability are increased thanks to this methodology, which enables it to detect minute variations in hand gestures and movements. Using the Keras and OpenCV libraries in conjunction with the Python programming language, we build a strong foundation for sign language recognition. In addition to streamlining the development process, these technological tools open the system to a larger developer and research community. In alignment with the research, the proposed work achieved an accuracy of 97% with overall precision, recall and F1 score of 99%.