Loading...
A Sensor-Based Approach to Sign Language Recognition Using Hidden Markov Model
Bayatmanesh, Saeid | 2013
646
Viewed
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 45388 (05)
- University: Sharif University of Technology
- Department: Electrical Engineering
- Advisor(s): Shamsollahi, Mohammad Bagher; Vossoughi, Gholamreza
- Abstract:
- Sign language is the first and most important communication way between hearing impaired community, but the biggest issue within them is simply that most of them can't effectively communicate with most hearing people. If sign language can be translated to text or speech automatically, deaf people will be able to communicate with all the others. Sign language contains more than six thousands signs, in which, deaf people make use of hands and sometimes facial expression to do that. So far, three main approaches have been used to recognize posture and position of hand: 1) vision-based: using images of 1-3 camera(s), based on image processing; 2) glove-based: using sensory glove(s) and motion tracker; and 3) hybrid approach. Position, orientation, movement and posture of the hands are the parameters that should be considered to completely recognize a sign. There are three different types of signs: 1) Isolated signs: which only one sign is performed in each time step; 2) Alphabet signs: used to express Specific names, 3) Continuous signs: in which signs are performed in sentence without artificial pause. In the last one, the problem of recognizing start and end of each sign should be considered. The Purpose of this project is to recognize sign language with the glove-based approach (data recorded by a glove and IMUs) and using Hidden Markov Model. Since there is no open access database for glove-based approach, data of 100 different signs of American Sign Language was recorded. Then appropriate features were extracted from raw data and a Hidden Markov Model was trained for each sign. After training, system can recognize isolated, alphabet and continuous signs with the accuracy of 97.75%, 98.08% and 93.33% respectively. This system can be generalized to other sign languages
- Keywords:
- Hidden Markov Model ; Sign Language Translation ; Continuous Signs Recognition ; Sign Language Recognition ; Sensory Glove ; Motion Tracker ; Inertial Measurement Unite (IMU)
- محتواي کتاب
- view