LSTM + Dynamic Time Warping for Gesture Sequence Recognition

Dynamic gesture recognition requires understanding temporal patterns in motion sequences. This post describes the approach published at IEEE GITCON 2026.

The Problem with Static Recognition

Most sign language recognition systems treat each frame independently. But dynamic signs — those requiring motion — need temporal context to be understood correctly.

LSTM for Temporal Modeling

Long Short-Term Memory networks excel at learning long-term dependencies in sequences. We used a bidirectional LSTM to process MediaPipe keypoint sequences.

Dynamic Time Warping for Matching

DTW allows comparison of two temporal sequences that may vary in speed or length — perfect for comparing gesture recordings from different signers.