WEIGHTED OPTICAL FLOW MODEL FOR VISEMES RECOGNITION IN SIGN LANGUAGE TUTORING SOFTWARE
Abstract
The problem of viseme recognition in video stream is considered. The problem arises as a part of sign language understanding and translation process. An approach that uses only dynamic viseme features is proposed. A weighted optical flow model is developed for modeling these features. Initialization, join, and distance operators are defined on weighted optical flow models for training and recognition purposes. The performance of the developed method was evaluated on a database of 119 visemas that belong to five classes. The recognition rate 69% was achieved for single user recognition.
Published
2019-02-26
How to Cite
Davydov, M., Nikol’skij, J., & Tykhanskyi, S. (2019). WEIGHTED OPTICAL FLOW MODEL FOR VISEMES RECOGNITION IN SIGN LANGUAGE TUTORING SOFTWARE. Electrotechnic and Computer Systems, (11(87), 132-141. Retrieved from https://eltecs.op.edu.ua/index.php/journal/article/view/1468
Section
Artificial Intellect Systems