Dynamic time wrapping based gesture recognition

Abstract

Sign language provides hearing and speech impaired people with an interface to communicate with society. Unfortunately most people do not understand sign language. For this, image processing and pattern recognition can provide with a vital tool to detect and translate sign language into vocal language. This work presents a method for detecting, understanding and translating sign language gestures to vocal language. Microsoft Kinect is the primary tool used to capture video stream of the user. This is achieved by getting skeleton frame from Kinect and then extracting joints of interest. The data obtained are normalized and a linked list of skeleton frame is maintained. The proposed method is capable of successfully detecting all gestures that do not involve finger movements. The proposed system has an accuracy of 91%.