Novel error correction-based key frame extraction technique for dynamic hand gesture recognition

Before languages came into existence, sign language was the mode of communication. For human-computer interaction, recognizing these sign languages is vital; thus, hand gesture recognition comes into play. With the advancement of technology and its vast applications, hand gesture recognition has become a common field of research. Gesture recognition has gained a lot of popularity due to its application in sign language detection for speech and hearing-impaired people. This paper presents a methodology for hand gesture recognition using a 3D convolutional neural network. The dataset used for this purpose is MINDS-Libras, a Brazilian sign language dataset. We propose a novel error correction-based key frame extraction technique that selects significant key frames for video summarization. The chosen key frames are preprocessed through the steps of the region of interest selection, background removal, segmentation, binarization, and resizing. The frames are given as input to the proposed three-dimensional convolutional neural network for the classification of hand gestures, which offers an accuracy of 98% and performs better than state-of-the-art techniques.