{"id":39378,"date":"2026-03-04T17:27:08","date_gmt":"2026-03-04T16:27:08","guid":{"rendered":"https:\/\/www.graviton.at\/letterswaplibrary\/working-on-a-low-cost-sign-language-recognition-system-for-hearing-impaired-students-need-advice-on-collecting-datasets\/"},"modified":"2026-03-04T17:27:08","modified_gmt":"2026-03-04T16:27:08","slug":"working-on-a-low-cost-sign-language-recognition-system-for-hearing-impaired-students-need-advice-on-collecting-datasets","status":"publish","type":"post","link":"https:\/\/www.graviton.at\/letterswaplibrary\/working-on-a-low-cost-sign-language-recognition-system-for-hearing-impaired-students-need-advice-on-collecting-datasets\/","title":{"rendered":"Working On A Low-cost Sign Language Recognition System For Hearing-impaired Students \u2014 Need Advice On Collecting Datasets"},"content":{"rendered":"<p><!-- SC_OFF --><\/p>\n<div class=\"md\">\n<p>Hi everyone,<\/p>\n<p>I&#8217;m a computer science student currently working on a project called \ud835\udc12\ud835\udc22\ud835\udc20\ud835\udc27\ud835\udc01\ud835\udc2b\ud835\udc22\ud835\udc1d\ud835\udc20\ud835\udc1e, an AI-powered accessible learning platform designed to improve classroom communication for hearing-impaired students.<\/p>\n<p>The main goal of the project is to build a \ud835\udc25\ud835\udc22\ud835\udc20\ud835\udc21\ud835\udc2d\ud835\udc30\ud835\udc1e\ud835\udc22\ud835\udc20\ud835\udc21\ud835\udc2d \ud835\udc2c\ud835\udc22\ud835\udc20\ud835\udc27 \ud835\udc25\ud835\udc1a\ud835\udc27\ud835\udc20\ud835\udc2e\ud835\udc1a\ud835\udc20\ud835\udc1e \ud835\udc2b\ud835\udc1e\ud835\udc1c\ud835\udc28\ud835\udc20\ud835\udc27\ud835\udc22\ud835\udc2d\ud835\udc22\ud835\udc28\ud835\udc27 \ud835\udc2c\ud835\udc32\ud835\udc2c\ud835\udc2d\ud835\udc1e\ud835\udc26 \ud835\udc2d\ud835\udc21\ud835\udc1a\ud835\udc2d \ud835\udc1c\ud835\udc1a\ud835\udc27 \ud835\udc2b\ud835\udc2e\ud835\udc27 \ud835\udc28\ud835\udc27 \ud835\udc25\ud835\udc28\ud835\udc30-\ud835\udc1c\ud835\udc28\ud835\udc2c\ud835\udc2d \ud835\udc1d\ud835\udc1e\ud835\udc2f\ud835\udc22\ud835\udc1c\ud835\udc1e\ud835\udc2c (\ud835\udc27\ud835\udc28\ud835\udc2b\ud835\udc26\ud835\udc1a\ud835\udc25 \ud835\udc25\ud835\udc1a\ud835\udc29\ud835\udc2d\ud835\udc28\ud835\udc29\ud835\udc2c \ud835\udc30\ud835\udc22\ud835\udc2d\ud835\udc21\ud835\udc28\ud835\udc2e\ud835\udc2d \ud835\udc06\ud835\udc0f\ud835\udc14\ud835\udc2c) so that it could realistically be deployed in schools.<\/p>\n<p>Current approach:<\/p>\n<p>&#8211; MediaPipe Holistic for hand + pose landmark extraction<\/p>\n<p>&#8211; Landmark normalization<\/p>\n<p>&#8211; Random Forest classifier for sign prediction<\/p>\n<p>&#8211; FastAPI backend + React frontend<\/p>\n<p>&#8211; Real-time webcam input<\/p>\n<p>The system currently supports \ud835\udc1b\ud835\udc1a\ud835\udc2c\ud835\udc22\ud835\udc1c \ud835\udc30\ud835\udc28\ud835\udc2b\ud835\udc1d-\ud835\udc25\ud835\udc1e\ud835\udc2f\ud835\udc1e\ud835\udc25 \ud835\udc2c\ud835\udc22\ud835\udc20\ud835\udc27 \ud835\udc1d\ud835\udc1e\ud835\udc2d\ud835\udc1e\ud835\udc1c\ud835\udc2d\ud835\udc22\ud835\udc28\ud835\udc27 and includes a \ud835\udc1c\ud835\udc25\ud835\udc1a\ud835\udc2c\ud835\udc2c\ud835\udc2b\ud835\udc28\ud835\udc28\ud835\udc26 \ud835\udc26\ud835\udc28\ud835\udc1d\ud835\udc1e \ud835\udc1f\ud835\udc28\ud835\udc2b \ud835\udc1b\ud835\udc22\ud835\udc1d\ud835\udc22\ud835\udc2b\ud835\udc1e\ud835\udc1c\ud835\udc2d\ud835\udc22\ud835\udc28\ud835\udc27\ud835\udc1a\ud835\udc25 \ud835\udc1c\ud835\udc28\ud835\udc26\ud835\udc26\ud835\udc2e\ud835\udc27\ud835\udc22\ud835\udc1c\ud835\udc1a\ud835\udc2d\ud835\udc22\ud835\udc28\ud835\udc27<\/p>\n<p>&#8211; Student signs \u2192 converted to text<\/p>\n<p>&#8211; Teacher speech \u2192 converted to live captions<\/p>\n<p>Right now the biggest limitation is \ud835\udc1d\ud835\udc1a\ud835\udc2d\ud835\udc1a\ud835\udc2c\ud835\udc1e\ud835\udc2d \ud835\udc2c\ud835\udc22\ud835\udc33\ud835\udc1e. I only have a small set of labeled sign images\/videos, which makes it difficult to expand vocabulary or experiment with temporal models.<\/p>\n<p>I&#8217;m looking for advice on a few things:<\/p>\n<ol>\n<li>\ud835\udc03\ud835\udc1a\ud835\udc2d\ud835\udc1a\ud835\udc2c\ud835\udc1e\ud835\udc2d\ud835\udc2c \ud835\udc1f\ud835\udc28\ud835\udc2b \ud835\udc08\ud835\udc27\ud835\udc1d\ud835\udc22\ud835\udc1a\ud835\udc27 \ud835\udc12\ud835\udc22\ud835\udc20\ud835\udc27 \ud835\udc0b\ud835\udc1a\ud835\udc27\ud835\udc20\ud835\udc2e\ud835\udc1a\ud835\udc20\ud835\udc1e (\ud835\udc08\ud835\udc12\ud835\udc0b) or similar landmark-based sign datasets.<\/li>\n<li>Best ways to \ud835\udc1c\ud835\udc28\ud835\udc25\ud835\udc25\ud835\udc1e\ud835\udc1c\ud835\udc2d \ud835\udc1a \ud835\udc2c\ud835\udc26\ud835\udc1a\ud835\udc25\ud835\udc25 \ud835\udc1b\ud835\udc2e\ud835\udc2d \ud835\udc2e\ud835\udc2c\ud835\udc1e\ud835\udc1f\ud835\udc2e\ud835\udc25 \ud835\udc1d\ud835\udc1a\ud835\udc2d\ud835\udc1a\ud835\udc2c\ud835\udc1e\ud835\udc2d for word-level or classroom-related signs.<\/li>\n<li>Suggestions for improving the model while keeping it \ud835\udc25\ud835\udc22\ud835\udc20\ud835\udc21\ud835\udc2d\ud835\udc30\ud835\udc1e\ud835\udc22\ud835\udc20\ud835\udc21\ud835\udc2d \ud835\udc1e\ud835\udc27\ud835\udc28\ud835\udc2e\ud835\udc20\ud835\udc21 \ud835\udc2d\ud835\udc28 \ud835\udc2b\ud835\udc2e\ud835\udc27 \ud835\udc28\ud835\udc27 \ud835\udc02\ud835\udc0f\ud835\udc14 \ud835\udc1d\ud835\udc1e\ud835\udc2f\ud835\udc22\ud835\udc1c\ud835\udc1e\ud835\udc2c.<\/li>\n<li>Any feedback on the system design or architecture.<\/li>\n<\/ol>\n<p>Eventually I\u2019d like to extend it toward \ud835\udc2c\ud835\udc1e\ud835\udc2a\ud835\udc2e\ud835\udc1e\ud835\udc27\ud835\udc2d\ud835\udc22\ud835\udc1a\ud835\udc25 \ud835\udc30\ud835\udc28\ud835\udc2b\ud835\udc1d \ud835\udc1d\ud835\udc1e\ud835\udc2d\ud835\udc1e\ud835\udc1c\ud835\udc2d\ud835\udc22\ud835\udc28\ud835\udc27 \ud835\udc28\ud835\udc2b \ud835\udc2c\ud835\udc22\ud835\udc26\ud835\udc29\ud835\udc25\ud835\udc1e \ud835\udc2c\ud835\udc1e\ud835\udc27\ud835\udc2d\ud835\udc1e\ud835\udc27\ud835\udc1c\ud835\udc1e-\ud835\udc25\ud835\udc1e\ud835\udc2f\ud835\udc1e\ud835\udc25 \ud835\udc22\ud835\udc27\ud835\udc2d\ud835\udc1e\ud835\udc2b\ud835\udc1a\ud835\udc1c\ud835\udc2d\ud835\udc22\ud835\udc28\ud835\udc27, but still keep it deployable on low-resource hardware. Currently this is done by the react side like when users sign it stores the sequence of words.<\/p>\n<p>If anyone has worked on sign language recognition, accessibility tools, or dataset collection, I\u2019d really appreciate your suggestions.<\/p>\n<p>Thanks<\/p>\n<\/div>\n<p><!-- SC_ON -->   submitted by   <a href=\"https:\/\/www.reddit.com\/user\/Agile_Commission1099\"> \/u\/Agile_Commission1099 <\/a> <br \/> <span><a href=\"https:\/\/www.reddit.com\/r\/datasets\/comments\/1rkpqhb\/working_on_a_lowcost_sign_language_recognition\/\">[link]<\/a><\/span>   <span><a href=\"https:\/\/www.reddit.com\/r\/datasets\/comments\/1rkpqhb\/working_on_a_lowcost_sign_language_recognition\/\">[comments]<\/a><\/span><\/p><div class='watch-action'><div class='watch-position align-right'><div class='action-like'><a class='lbg-style1 like-39378 jlk' href='javascript:void(0)' data-task='like' data-post_id='39378' data-nonce='9941108d62' rel='nofollow'><img class='wti-pixel' src='https:\/\/www.graviton.at\/letterswaplibrary\/wp-content\/plugins\/wti-like-post\/images\/pixel.gif' title='Like' \/><span class='lc-39378 lc'>0<\/span><\/a><\/div><\/div> <div class='status-39378 status align-right'><\/div><\/div><div class='wti-clear'><\/div>","protected":false},"excerpt":{"rendered":"<p>Hi everyone, I&#8217;m a computer science student currently working on a project called \ud835\udc12\ud835\udc22\ud835\udc20\ud835\udc27\ud835\udc01\ud835\udc2b\ud835\udc22\ud835\udc1d\ud835\udc20\ud835\udc1e, an AI-powered accessible&#8230;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[85],"tags":[],"class_list":["post-39378","post","type-post","status-publish","format-standard","hentry","category-datatards","wpcat-85-id"],"_links":{"self":[{"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/posts\/39378","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/comments?post=39378"}],"version-history":[{"count":0,"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/posts\/39378\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/media?parent=39378"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/categories?post=39378"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/tags?post=39378"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}