Filipino Sign Language Recognition for Beginners using Kinect

Abstract

There exists a shortcoming called communication gap between deaf people and people who can hear. Most people use verbal language as a form of communication while deaf people use non-verbal communication called sign language. Several studies were already conducted in the field of signal processing focusing on sign language recognition. This study covers Filipino sign language and used Kinect V2. The proponents of this research analyzed the performance between two different methods. In the first method, the location of the joints of interest are tracked via the Cartesian coordinates of the joints while the second method tracks the location of the joints via spherical coordinates that includes normalization. Normalization is applied because the size and position of the user from the Kinect V2 may vary resulting to poor recognition. Both of these methods used the Dynamic Time Warping algorithm and the Support Vector Machine individually. In total there are four results: results for the Spherical-DTW, Spherical-SVM, Cartesian-DTW and Cartesian-SVM. The scope of this study includes basic Filipino words provided by sign language experts wherein the hands are in open palm shape. Fingers and facial expression are not considered therefore, these gaps can be an opportunity for future studies. The proposed method in this study reached a peak accuracy of 95.00%, recall of 95.00%, and precision of 95.89%