Two-Stage Hand Gesture Recognition based on Hand Keypoints Localization

Hand gesture is an important component of non-verbal communication, and the appropriate categorization of the gestures is quintessential for fruitful communication. Hand gestures are used in many human-computer interfaces for their natural and simplistic contactless way of conveying instruction to the interface. However, the recognition of hand gestures is complicated by numerous factors. This paper addresses a few issues by proposing a two-stage recognition framework that uses a hand joint localization technique. Firstly, the proposed method predicts hand keypoints that localize the region of interest by encompassing the hand region through a bounding box. Subsequently, this region of interest is used in gesture recognizing. The proposed work uses only one input modality-RGB image and performs phenomenally despite background clutter and illumination variation.