Research Build

Sign Language Recognition with Graph Neural Networks

Built a computer vision pipeline that combines pose and hand keypoints, graph attention, and temporal modeling to classify sign language gestures from video.

Year2022

Impact

Explored how spatio-temporal graph representations can improve gesture understanding beyond frame-only visual models.

Problem

Gesture understanding from video needs both precise spatial reasoning across joints and temporal reasoning across frame sequences.

Approach

I represented body and hand motion as spatio-temporal graphs, then combined graph attention with LSTM-based sequence modeling for gesture classification.

Outcome

The system demonstrated a strong research direction for sign language recognition using graph-based visual representations instead of standard frame classifiers alone.