A Review of the Embedded Deployment of Sign Language Detection Models: Metrics, Tradeoffs and Perspectives

Abstract

Sign language is the unique means of communication between deaf-mute people and society members. Unfortunately, the lack of sign language learning in many cultures along with the challenges associated with understanding it, leads to the isolation of deaf-mute individuals from their community. In response to this issue, recent years have witnessed a notable surge in research on sign language technologies, aimed at bridging the communication gap and fostering greater inclusivity. Leveraging recent advances in AI algorithms, embedded hardware, internet of things and emergence of Tiny Machine learning have paved the way for a revolution in the use of embedded systems for sign language recognition. Despite these significant advancements, a key remaining challenge is achieving the trade-off between high performance and resource constraints of embedded devices. To this end, the focus of this review is to provide a comprehensive overview of the complete pipeline from data collection to deployment of deep or machine learning models on embedded edge devices. It also offers a comparative of distinct hardware and framework tools used in these embedded applications of sign language recognition published from 2020 to 2025. Literature findings of this review show the possibility of designing an efficient embedded application which can recognize sign language in real-time. Furthermore, this study presents the future directions in order to accelerate the commercial deployment of these systems.