Scholarlink

Sign-language video search systems often rely on gloss labels, manual annotations, or closed vocabularies, limiting scalability in real-world use. We present a fully automated, annotation-free American Sign Language (ASL) retrieval system for looking up the meaning of unknown signs. Using MediaPipe, we extract hand and upper body pose landmarks from pre-segmented sign videos, compute temporal sequences of inter-segment angles, and measure similarity with Dynamic Time Warping (DTW). Given a query clip, the system returns the top-50 nearest matches from a precomputed sign database. On a 1,113 sign lexicon, the correct sign appears in the top-5 results for 66% of queries and in the top-10 for 75%. These results indicate the potential for lightweight, annotation-free ASL lookup tools suited to learners and assistive applications in low-resource environments.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Angle Based DTW for Large Vocabulary Sign Retrieval

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yanli Dong

摘要

Sign-language video search systems often rely on gloss labels, manual annotations, or closed vocabularies, limiting scalability in real-world use. We present a fully automated, annotation-free American Sign Language (ASL) retrieval system for looking up the meaning of unknown signs. Using MediaPipe, we extract hand and upper body pose landmarks from pre-segmented sign videos, compute temporal sequences of inter-segment angles, and measure similarity with Dynamic Time Warping (DTW). Given a query clip, the system returns the top-50 nearest matches from a precomputed sign database. On a 1,113 sign lexicon, the correct sign appears in the top-5 results for 66% of queries and in the top-10 for 75%. These results indicate the potential for lightweight, annotation-free ASL lookup tools suited to learners and assistive applications in low-resource environments.