Recognition of Sign Language Alphabets and Numbers based on Hand Kinematics using A Data Glove

Abstract

This paper reports real-time recognition of Indian and American sign language alphabets and numbers based on hand kinematics assessment. The finger and wrist joint angles were acquired using an indigenously developed data glove. The data set was for single handed Indian sign language alphabets (C, I, J, L, O, U, Y, W), American sign language alphabets (A to Z) and sign numbers (0 to 9). The data were pre-processed through a moving average filter and standardized feature scaling methods. The glove was able to measure the finger joint angles with an accuracy±standard deviation for metacarpophalangeal (MCP) joint±2.14°, proximal inter phalangeal (PIP) joint± 1.73° and distal inter phalangeal (DIP) joint± 1.49°, during flexion/extension and abduction/adduction movements. A radial basis function kernel support vector machine with 10-fold cross validation was used for recognition. An average recognition rate of 96.7% was achieved. Using a label matching and speech data base, the recognized alphabets and numbers were translated into speech.