Understanding human interaction in RGB-D videos

Abstract

In this report, a human hand gesture recognition system is proposed. The system can

understand both static and dynamic human hand gestures. So far, the system is able to

recognize 9 static hand gestures: numbers from one to nine; and 1 dynamic hand gesture:

number ten. In the system implementation, a right-hand CyberGlove II is used to get the

accurate and stable hand joints information for the static hand gesture recognition. Based on

the results of static classification, together with the hand joint motion information from

Microsoft Kinect, dynamic hand gestures can be classified. In addition, and effective and fast

human hand gesture recognition algorithm is proposed to manage the data from sensors and

achieve classification results in real time. To verify the effectiveness of the system, a human

hand gesture sample dataset containing 250 samples collected from 5 people of difference

body sizes is constructed. The testing results show that the algorithm is able to understand

human hand gestures accurately and fast.