Garbage Model Formulation for Sign Language Spotting with Conditional Random Fields(Internationa Session 7)

概要

論文の詳細を見る
Automatic recognition and spotting of sign language from continuous signing present a number of challenges. The signs appear within a continuous gesture stream, interspersed with transitional movements between sign (communicative) and non-sign (non-communicative) movements. Many previous sign spotting methods employed a fixed threshold that best discriminates signs and non-sign movements. However, it is difficult to select one fixed threshold that works well for all signs. In this paper, a novel method for designing a garbage model in conditional random field (CRF) that perform an adaptive threshold for distinguishing between sign and non-sign movements by augmenting the CRF with one additional label is proposed. A short sign detector, a hand appearance-based sign verification method, and a subsign reasoning method are included to further improve spotting accuracy. The short sign detector models signs that tend to have shorter than normal duration. The hand appearance-based sign verification method overcomes the ambiguity among signs that exhibits similar overall hand movements, but differ in hand shape. Finally, the subsign reasoning method avoids premature detection of a sign that can be made by parts of longer sign movements. Experiments demonstrate that our system can detect signs from continuous data with an 88% spotting rate and can recognize signs from isolated data with a 94% recognition rate, versus 61% spotting rate and 85% recognition rate respectively for CRF without a garbage label.
社団法人電子情報通信学会の論文
2007-10-18

Garbage Model Formulation for Sign Language Spotting with Conditional Random Fields(Internationa Session 7)

スポンサーリンク

概要

著者

関連論文

スポンサーリンク