ISL gesture recognition using multiple feature fusion

Communication is an important part of our day to day lives. It is an essential means of conveying our emotions through gestures as well as verbally. However, communication and recognition of gestures becomes a problem for humans with speaking and hearing impairment. For this very purpose sign language is used. Indian Sign Language (ISL) is resourceful in India for helping such people. The paper proposes a novel technique to identify gestures defined for the English alphabets listed in the Indian Sign Language. The proposed techniques relies on multiple representations namely HOG, GIST and BSIF. A random forest classifier is used for classifying different gestures with the aid of the combined feature vector. The results of the proposed techniques were tested on a ISL hand gesture database and were compared with the existing solutions. The technique outperformed the existing solutions resulting in an accuracy of 92.20%.