Dynamic gesture recognition and human-computer interaction

In this paper, the recognition algorithm of dynamic gestures, which based on depth information, is proposed. A new human-computer interaction controlled by dynamic gestures is also achieved. The proposed algorithm extracts Histogram of Oriented Gradients (HOG) and weighted Hu invariant moments from RGB and depth map of the interested region of gestures respectively, and fuses weighted Hu invariant moments into HOG. Then, we establish Support Vector Machine model (SVM) for static hand gestures recognition. Base on this, an HMM toolbox supporting multi-dimensional continuous data input is adopted to do the training and dynamic gesture recognition. Finally, the visual simulation software Vega Prime is applied to simulate a virtual UAV flight controlled by dynamic gestures. Experimental results demonstrate that the proposed algorithm is effective and accurate in complex background and practical real-time applications.