Team · Team Project | Video Understanding and Inclusive AI
Continuous Sign Language Translation
An inclusive-AI video understanding pipeline that translates continuous sign-language gestures into coherent, usable text output.
My contribution
I contributed to the vision-to-language pipeline: sequence handling, model experimentation, and keeping accessibility goals concrete in the technical design.
Python · Computer vision · Deep learning · Sequence modelling · NLP · Video understanding · Keypoint features
Problem
Isolated gesture classifiers fail in real-world communication because natural sign language relies on continuous temporal transitions, co-articulation, and grammar rather than static hand poses.
Solution & architecture
Engineered an end-to-end spatio-temporal deep learning system capable of extracting frame-level visual/keypoint features and mapping continuous video sequences to natural language tokens.
Sequence modeling
Frame sampling → spatial feature extraction → temporal sequence alignment/decoding → post-processed text generation. Key challenges included variable signing speeds, motion blur, and sequence boundary segmentation without manual pauses.
My contribution
Team contribution across pipeline design and experimentation for continuous sign-language translation.
Impact
Advances accessibility tooling by bridging continuous visual sign streams with readable text interfaces — accessibility treated as an engineering problem, not a marketing label.