Research map / Large language models / Language, safety and interpretability
A Wireframe-Based Approach for Classifying and Acquiring Proficiency in the American Sign Language (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Converts ASL hand-gesture images into hand wireframes via TensorFlow hand landmarks, trains a small CNN and a RandomForest on joint angles to classify signs (94% and 97% test accuracy), and proposes angle-deviation feedback for sign tutoring.
Problem
Recognizing American Sign Language gestures from raw images is prone to biases from skin tone, texture, and finger thickness, and learners lack targeted feedback on hand-gesture errors.
Approach
Replace raw images with landmark-based wireframes (rendered on black backgrounds) and further distill them into finger joint angles, classified with a RandomForest, and compare a learner's joint angles to average reference angles per sign to give targeted feedback.
Key results
- Sequential 8-layer deep network on wireframe images reaches 94% test accuracy.
- RandomForest on joint angles improves test accuracy to 97%.
- Wireframe approach is claimed to be more memory efficient than the deep network.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use joint angles and RandomForests instead of the deep network on wireframes | ASL sign classification · Curated ASL test set | accuracy: 94% -> 97% | Helped |
Limitations
- Only still-image letters and digits; J and Z excluded because they require motion.
- Feedback is numerical deviation only; visual overlay guidance is planned but not implemented.
- Small test split (5%) and no comparison with raw-image baselines or external benchmarks.
Experimental setup
Datasets: Kaggle ASL alphabet dataset
Benchmarks: curated ASL wireframe test split
Metrics: accuracy
Closest papers by meaning
- FSboard: Over 3 Million Characters of ASL Fingerspelling Collected via Smartphones (CVPR 2025 · similarity 0.87)
- VSNet: Focusing on the Linguistic Characteristics of Sign Language (CVPR 2025 · similarity 0.86)
- KD-MSLRT: Lightweight Sign Language Recognition Model Based on Mediapipe and 3D to 1D Knowledge Distillation (AAAI 2025 · similarity 0.85)
- TCNet: Continuous Sign Language Recognition from Trajectories and Correlated Regions (AAAI 2024 · similarity 0.84)
- YouTube-SL-25: A Large-Scale, Open-Domain Multilingual Sign Language Parallel Corpus (ICLR 2025 · similarity 0.84)
- SignGraph: A Sign Sequence is Worth Graphs of Nodes (CVPR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.