Isolated Sign Language Recognition with Segmentation and Pose Estimation

TL;DR

This paper presents a novel ISLR model that combines pose estimation, segmentation, and a ResNet-Transformer backbone to improve recognition accuracy and efficiency for American Sign Language videos, addressing data scarcity and signer variability.

Contribution

It introduces an integrated approach that reduces computational costs and enhances robustness to signer differences in isolated sign language recognition.

Findings

Achieves competitive accuracy with lower computational requirements.

Effectively handles signer variability in sign language videos.

Demonstrates robustness through integrated pose and segmentation modules.

Abstract

The recent surge in large language models has automated translations of spoken and written languages. However, these advances remain largely inaccessible to American Sign Language (ASL) users, whose language relies on complex visual cues. Isolated sign language recognition (ISLR) - the task of classifying videos of individual signs - can help bridge this gap but is currently limited by scarce per-sign data, high signer variability, and substantial computational costs. We propose a model for ISLR that reduces computational requirements while maintaining robustness to signer variation. Our approach integrates (i) a pose estimation pipeline to extract hand and face joint coordinates, (ii) a segmentation module that isolates relevant information, and (iii) a ResNet-Transformer backbone to jointly model spatial and temporal dependencies.