An Effective Approach for Reducing Time and Memory Requirements for Sign Language Detection

Le résumé fourni par la source

Sign language serves as a crucial means of communication for individuals who are hard of hearing or deaf, but it is often misunderstood or unrecognized by the hearing population, leading to communication barriers and social exclusion. To address this issue, this research work introduces an approach that utilizes Mediapipe data and neural networks to accurately detect individual sign language gestures. By preprocessing the data, the size is significantly reduced by focusing solely on landmark points rather than entire videos. Furthermore, the preprocessing techniques decrease the training data size by approximately four times. The proposed method employs simple fully connected layers, avoiding computationally intensive CNN or LSTM architectures typically used in existing sign language models. Consequently, the training time is significantly reduced, from 7 hours to just 22 minutes, making the process more efficient and accessible. The effectiveness of the approach is evaluated across a range of 10 to 250 signs, surpassing the limited number of signs explored in most existing research. The achieved accuracies range from 84% for 10 signs to 73% for 250 signs. These results show the efficacy of the proposed approach in facilitating daily communication for individuals with hearing impairments. By employing Mediapipe data and leveraging neural networks with simplified architectures, this approach offers a promising solution for accurate sign language gesture detection, providing an avenue for improved understanding and inclusion between the deaf and hearing communities.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.