Hand Gesture Recognition with Batch and Reinforcement Learning

In this paper, we present a system for real-time recognition of user-defined static hand gestures captured via a traditional web camera. We use SURF descriptors to get the bag-of-visual-words features of the user’s hand, and use these features to train a multi-class supervised learning model. We choose the best learning model from (SVM, Neural Networks, Decision Trees, and Random Forests) and the best model parameters using hyper-parameter optimization algorithm. During test time, we use these bag-of-visual words features to predict the users hand gestures. The user has the ability to give positive or negative feedback for every prediction to the system, and the system updates itself during test time for better accuracy.