Ensembled Neural Network for Static Hand Gesture Recognition

Abstract

Sign language was originally created to fulfil the gap of communication between the speech impaired. However, with recent advances, we can see the applications of sign language in a variety of different fields such as automated vehicle movements, assistant systems, human robot interaction. In this paper, the main focus is towards creating a system that is highly efficient in detecting the hand sign gestures. To achieve these results, numerous deep learning techniques and advanced models have been used in this paper. The models were trained on ASL (American Sign language) dataset in which the highest validation accuracy was 99.56% achieved with Resnet 50 architecture. Other pre-trained models used also received similar validation accuracies. A combined voting classifier is also implemented to achieve the maximum accuracy. All the models were finally tested on the HGM -4 dataset which is a totally new dataset with a different set of images and our models obtained decent results here also as shown in Table 1.