OFFLINE CANDIDATE HAND GESTURE SELECTION AND TRAJECTORY DETERMINATION FOR CONTINUOUS ETHIOPIAN SIGN LANGUAGE
Abadi Tsegay · 2011
A lot of effort has been invested in developing alphabet recognition and continuous sign language translation systems for many sign languages around the world. In this regard, little attention has been given to Ethiopian sign language (EthSL). However, an Ethiopian Manual Alphabet (EMA) recognition system has been developed in 2010. For a recognition system that can recognize continuous gestures from video which can be used as a translation system, a methodology that selects candidate gestures from sequence of video frames and determines hand movement trajectories is required. In this paper, a system that extracts candidate gestures for EMA and determines hand movement trajectories is proposed. The system has two separate parts namely Candidate Gesture Selection (CGS) and Hand Movement Trajectory Determination (HMTD). The CGS combines two metrics namely speed profile of continuous gestures for block division (BD) and Modified Hausdorff Distance (MHD) measure for gesture shape comparison and has an accuracy of 80.72%. The HMTD is done by considering the centroid of each hand gesture from frame to frame and using angle history, x-direction and y-direction of lines between successive centroids. A qualitative evaluation of the CGS is found to be 94.81%. The HMTD has an accuracy of 88.31%. The overall system performance is 71.88%.