Анализ мануальных компонентов украинской жестовой речи с использованием системы дополнительных маркеров

Abstract

Рассмотрено создание информационной технологии распознавания отдельных жестов украинского жестового языка, полученных посредством технологии Motion Capture. Методами исследования служат алгоритмы динамической свертки временной последовательности.

Розглянуто створення інформаційної технології розпізнавання окремих жестів української жестової мови, отриманих за технологією Motion Capture. Методами дослідження слугують алгоритми динамічної згортки часової послідовності.

Introduction. Gesture recognition is an actual problem concerning the interaction between a user and the computers. Although some prototypes of foreign sign language recognition systems have been developed and already used in computer vision, Ukrainian sign language recognition still remains the problem. Purpose. The aim is to develop a universal recognition technology for single Ukrainian signs captured with Motion Capture technology. The objects of the study are data bases of single signs, presented as time sequences of coordinates of motion components. The research methods cover the dynamic time warping algorithms. Results. Data model for gesture recognition in small Ukrainian sign language dictionaries is investigated. A DTW (Dynamic Time Warping) algorithm is used. The modified WDTW (Weighted Dynamic Time Warping) using weights that the more an individual marker for moving it in space during the whole demonstration gesture are proposed. The criterion of success algorithm performed relative number of correctly identified samples from a test one to their total number. Both algorithms are demonstrated a full recognition accuracy on the basis of 139 unique gestures. For WDTW algorithm was used normalization, and therefore the results have a high level of generalization. Conclusions. The WDTW algorithm to create a universal information technology of identification the Ukrainian sign language gestures is used. The gestures in the form of time series calculated using the technology Motion Capture are presented. This frame rate of 10 fps is sufficient for correct identification sign in a limited vocabulary