PSL-CFRT: A Benchmark Multi-View Dataset for Continuous Pakistani Sign Language Fingerspelling Recognition and Translation

Fingerspelling, an essential component of sign language, utilizes distinct hand configurations for each letter of a written language to convey proper names, technical terms, addresses, numerical values, and other uncommon words. This intricate form of communication is challenging for the general community, creating a communication barrier between the deaf and hard-of-hearing (DHH) community and those unfamiliar with it. To bridge this gap, this paper introduces a novel multi-view dataset specifically designed for continuous fingerspelling recognition and translation in Pakistani Sign Language (PSL). This dataset is unique in its inclusion of diverse fingerspelling variations, enhancing its applicability to real-world scenarios. In addition to introducing the dataset, an object-detection-based method is proposed for PSL continuous fingerspelling recognition (PSL-CFR), marking a significant advancement in continuous PSL finger-spelling recognition technology, achieving a baseline accuracy of 63.16%.