Enabling Real-time Sign Language Translation on Mobile Platforms with On-board Depth Cameras
summary
SUGO is a depth-only, on-device sign-to-text system that runs real-time on a smartphone using the front depth camera and no cloud. It uses data augmentation and image-processing schemes to emphasize gestures, a lightweight 3-dimensional convolutional neural network for frame-sequence classification, and a dataset from 20 individuals covering 50 Korean Sign Language words (~5,000 gestures plus in-the-wild examples). Evaluation reports up to 91% accuracy with low latency and mobile resource use, robust to motion and varied lighting.
AI-written summary · about
related talks
10mSignGlass: First-Person View Comprehensive and Generalizable ASL Translation Using Wearable Glass
30mBuilding a Sign-to-Speech Prototype with TensorFlow, Pytorch, and DeepStack
09mBeyond Captions A Practical Pipeline for Live AI-Generated Sign Language Avatars
45mWhat if AI were the solution to understanding sign language?
27mSign Language Recognition Improvements Since the London Keynote
08mIndian Sign Language Recognition (ISLAR)
35mHow deep learning can help interpret American Sign Language
15mKeynote: Empowering Accessibility Through Kubernetes: The Future of Real-Time Sign Language Interpretation