Real Tme Recognition of American Sign Language Based on Hand Posture Using RGB-D Camera

Please enable javascript in your browser.

Iranmanesh, Mohammad | 2015

487 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 47770 (19)
University: Sharif University of Technology
Department: Computer Engineering
Advisor(s): Jamzad, Mansour
Abstract:
Sign Language Recognition using cameras is an alternative way of human communication with personal computers. These systems don’t need to use keyboard. They are used when touching keyboard is not possible or for what ever reason, when we don’t want to use keyboard. In this problem the person shows one of the alphabets with his hand and we capture and process that in real-time to recognize the letter. The problem of previous works in this area are the decreasing rate of detection in condition where the angle of sign and viewer changes. The purpose of our work is getting better result by using depth image from Kinect. We use Pugeault and Bowden dataset (that has fingerspelling sign in multiple view) to achieve this goal. We could achieve the rate of 97/38% with using feature of FEMD for clustering and using best features (ESF Discriptors, Gabor Filter or Fourier Discriptors) and best classifier (SVM or RF) for classification that were choosen and trained with specifications of samples of each clusters. Our result is 97/38% that is very close to the state-of-the-art result of this dataset that is 97/8% . This algorithm can be used as real-time process
Keywords:
RGB-D Camera ; Human Computer Interaction (HCI) ; Real Time System ; Hand ; Kinect Sensor ; Hand Gestures ; American Sign Language

No TOC