Implementation and Performance Analysis of Indian Sign Language Recognition on various Computing Devices

Sign Language is a language that uses hand and facial gestures for communication. Its automated recognition is important as it would aid deaf and mute community since they use Sign Language to communicate. In this work an image data set of Indian Sign Language having 36 classes of commonly used words is presented. Three deep learning models VGG16, VGG19 and InceptionV3 were used for classification of the images in the data set achieving accuracy as high as 99.7852%. Three computing devices mobile CPU, High Performance Computing (HPC) machine, and Intel Neural Compute Stick 2 (Intel NCS 2) were used for measuring inference time with the least time recorded to be 77.2819 seconds. Comparative study of the performance of the deep learning models and computing devices has been presented in this paper.