Designing a Real-Time ASL Auto-Recognition Tool Using Deep Learning

American Sign Language comprises more than 500 signs and gestures. Assessing sign-language for each individual learner is often time-consuming, especially when instructors assess learners' work regularly. In this project, we describe the design of an ASL auto-recognition tool, which can evaluate students' signed videos automatically and hence help with self-evaluation. This tool uses deep learning methods to recognize ASL gestures and provides immediate feedback to ASL learners. The results suggest that the accuracy of the model is acceptable in conditions with adequate lighting. Hence, the tool can potentially be used to support instructors with sign language assessment.