Vision-based Hand Gesture Recognition for Indian Sign Language Using Convolution Neural Network
Jayesh Gangrade, Jyoti Bharti
Abstract
Hearing-impaired people can interact with other people through sign language. The proposed system tears down the communication barrier between Hard of hearing (HoH) community and those who do not know their sign language. In this paper, we have developed an algorithm to detect and segment the hand region from a depth image using the Microsoft Kinect sensor. The proposed algorithm works well in the cluttered environment, e.g. skin color background and hand overlaps the face. Convolution Neural networks (CNN) are applied to automatically construct features from Indian sign language (ISL) signs. These features are invariant to rotation and scaling. The proposed system recognizes gestures accurately up to 99.3%.
Topics & Concepts
Sign languageGestureComputer scienceGesture recognitionArtificial intelligenceConvolutional neural networkComputer visionSign (mathematics)Face (sociological concept)Convolution (computer science)Invariant (physics)Speech recognitionConstruct (python library)American Sign LanguageArtificial neural networkMathematicsSociologyMathematical physicsPhilosophySocial scienceMathematical analysisProgramming languageLinguisticsHand Gesture Recognition SystemsHearing Impairment and CommunicationGaze Tracking and Assistive Technology