Deep Learning Approach for Keypoint-Based Bangla Word Sign Detection for Videos

Abstract

This research delves into the realm of Bangla sign language recognition, focusing on developing a robust system for interpreting and analyzing gestures depicted in videos. Employing a keypoint based approach and leveraging deep learning technology, specifically a two-layer LSTM architecture, our methodology aims to interpret Bangla word signs accurately. The project's foundation lies in a meticulously curated custom dataset featuring 51 unique signs captured by two proficient signers, ensuring comprehensive coverage of articulation, speed, and style variations. Our system's tailored approach acknowledges a bridge to communication gaps and enhances accessibility for the Bangla-speaking Deaf community. Beyond technical advancements, this research aims to elevate societal awareness and foster employment opportunities for the Deaf population, ultimately contributing to a more inclusive world. The outcomes of this project have the potential to not only revolutionize Bangla sign language recognition but also pave the way for broader applications in accessibility and advancements in deep learning technology for diverse sign languages worldwide.