Deep learning for sign language recognition

A group project applying deep learning to video classification on the World-Level American Sign Language (WLASL) dataset, where each clip shows a single signed word.

My contribution was a convolutional approach that sidesteps video modelling entirely. Rather than processing the clip as a temporal sequence, it extracts 9 frames from the middle of the video and tiles them into a single 3×3 grid image — so the temporal structure of the sign becomes spatial structure in one static picture, and the problem reduces to ordinary image classification.

On top of that representation I built:

• Data augmentation suited to video: zoom, speed-up, and horizontal flip.
• A two-block CNN, each block being convolution → batch normalisation (batches of 10 images) → average pooling, feeding a dense multi-layer perceptron for the final classification.

Code: github.com/TomGiaco/Deep-Learning-to-predict-Sign-Language