45 Character Bangla Sign Language Indore Dataset

Sign languages are vital for communication within the deaf and hard-of-hearing community. This paper introduces a novel dataset designed explicitly for the recognition of Bangla Sign Language (BdSL). Our comprehensive dataset comprises 45 distinct classes, encompassing the 10 Bangla digits, 32 Bangla characters, and 3 special characters. Initially, 200 images were collected for each class, resulting in a robust base dataset. To enhance the dataset's diversity and mitigate overfitting, we employed data augmentation techniques, expanding each class to 800 images. The augmented dataset was then meticulously split into training and testing sets, with the training portion further divided to create a dedicated validation set. This structured approach ensures a thorough evaluation of machine learning models trained on this dataset. This BdSL dataset is intended to serve as a valuable resource for researchers and developers working on sign language recognition systems, fostering advancements in human-computer interaction and promoting inclusivity for the Bangla-speaking deaf community.

手语是聋人与重听群体开展沟通的核心媒介。本文提出了一款专为孟加拉语手语（Bangla Sign Language, BdSL）识别任务打造的新型数据集。该完整数据集涵盖45个独立类别，包含10个孟加拉语数字、32个孟加拉语字母以及3个特殊字符。最初我们为每个类别采集200张图像，由此构建起基础的高质量数据集。为提升数据集多样性并缓解过拟合问题，我们采用数据增强技术，将每个类别的样本量扩充至800张。随后我们对增强后的数据集进行精细划分，分为训练集与测试集，并将训练集进一步拆分出专属验证集。此结构化流程可确保基于该数据集训练的机器学习模型得到充分评估。本孟加拉语手语数据集旨在为从事手语识别系统研发的研究人员与开发者提供宝贵资源，推动人机交互领域的技术进步，并助力孟加拉语聋人群体的社会包容。