Static Video Based Visual-Verbal Exemplar for Recognizing Gestures of Indian Sign Language

The paper presents a system developed for recognizing gestures of Indian sign language from images of gestures. The proposed system is based on Elliptical Fourier descriptors and neural networks used for gesture pattern recognition. Unlike the systems proposed by other researchers such as using a radio frequency or colored gloves to achieve the recognition our system does not impose any such constraints. Features are extracted from the videos of signers using elliptical Fourier descriptors and principal component analysis which greatly reduces the size of the feature vector. Neural networks error back propagation algorithm is used to recognize gestures Indian sign language. The system converts the recognized gesture in to voice and text messages. The system was implemented with 440 sample videos of gestures of alphanumeric characters and words with a maximum of 5 videos per gesture. Experimental results show that the neural network is able to recognize gestures and convert them to voice messages with an accuracy of 92.52%.