Static Hand Gesture Recognition for Human Robot Interaction

Human-robot interaction is making a robot understanding human action, working and sharing the same space with the human. To achieve this, the communication between human and robot must be effective. The most used ways in this communication are the vocal and the body gestures such as full body actions, hand and arm gestures or head and facial gestures. Hand gestures are used as a natural and effective way of communicating between human and robot. The difference in hand size and posture, light variation and background complexity makes the hand gesture recognition to be a challenging issue. Different algorithms have been proposed and they gave better results. Yet, some problems such as poor computational scalability, trivial human intervenes and slow speed learning are still appearing in this field. In this paper, these issues are solved using a combination of three features extraction methods: Haralick texture, Hu moments, and color histogram and extreme learning machine (ELM) method for classification. The ELM results were compared to that of K-Nearest Neighbors, Random Forest Classifier, Linear Discriminant Analysis, Convolution Neural Networks and these experiment was evaluated on National University of Singapore (NUS) dataset II. ELM performed better than any of the above algorithms with an accuracy of 98.7 and time used of 109.7 s which is the proof of the satisfactory of the model.