Using Machine Learning and Deep Learning for Static and Dynamic Hand Gesture Recognition

Human-computer interaction aims at facilitating interaction between the hearing-impaired people and the hearing community. It focuses more on the gestural interfaces that require hand movements, which need complex algorithms to be identified. The objective of this paper is to recognize and identify static and dynamic hand gestures by using deep neural network technique, which is based on ResNet18 and a machine learning technique that is based on LGBPHS descriptor. In addition, one-against-all Support Vector Machine and the k-nearest neighbor are applied to classify the feature hand gestures obtained by these two different systems. The findings showed that ResNet18 led to higher hand recognition accuracy than LGBPHS descriptor. Regarding static hand gestures, a result of 99.16% was achieved using ResNet18 on Jochen Triesch’s dataset, and a recognition rate of 90.66% was obtained on Arabic sign language dataset. For dynamic hand gestures, a recognition rate of 91.66% was found on Sebastien Marcel.