Honours projectsSample project
SignBridge
Recognises South African Sign Language fingerspelling in real time from a webcam and turns it into text.
- Category
- Accessibility
- Year
- 2024
- Student
- Naledi Dlamini
- Supervisor
- Dr Ruan Esterhuizen
Research question
Few tools support South African Sign Language, which leaves Deaf students relying on interpreters for everyday conversations.
Approach
Hand landmarks are tracked in the browser and a sequence model classifies fingerspelled letters, with a language model correcting the output.
Key features
- Runs in the browser with an ordinary webcam
- Live captions of fingerspelled words
- Practice mode with feedback on hand shapes
Evaluation
Accuracy was measured per letter and per word with signers who were not part of the training data.