A Smart Vision Based Single Handed Gesture Recognition system using deep neural networks

. The primary and expressive mode of human communications are gestures. Human can interact with machines using body postures and finger pointing. Advancements in human-computer interaction (HCI) has presented new innovations in technology making the users to communicate with computers in an instinctual manner. Evidences clearly state that future living space will be dominated by sensor-based devices and hence an efficient human-computer interfaces are required to exchange information. Hand gesture interfaces have been employed in multiple domains and has won social acceptance. System requirements for gesture recognition vary with the intended application areas. Responsiveness, Learnability, Cost and Accuracy are major drivers for success of hand gesture recognition systems. This paper suggest a HCI design that requires no wearable markers or gloves. A noninvasive vision based framework has been suggested for human-machine interface. Deep Neural Networks have provided promising results in vision based tasks. Convolutional Neural Networks (CNN) are claimed for image recognition problems as they learn features from images gradually and automatically. An optimal CNN architecture has been proposed to recognize single handed gestures. The images of hand gestures convey a numerical representation of ten digits. Image augmentation has been performed to increase the size of training data for deep learning. Depending on application, the interpretation of gesture can be customized. The classification performance has been analyzed with metrics reported by confusion matrix. The proposed architecture performs well both in training and testing reporting the accuracy of 98.2% and 96.2% respectively. Tuning the hyper parameter has improved test accuracy. Hand gesture recognition, Deep Learning, Convolutional Neural Network (CNN).