Identification of Odissi dance video using Kinect sensor
Sriparna Saha, Shreya Ghosh, Amit Konar, R. Janarthanan · 2013
This paper introduces an algorithm for identification of dance video by recognizing posture from each frame for the purpose of e-learning. We are taking Indian classical dance `Odissi' as the input. The twenty videos `Chowkh' and `Tribhangi' of `Odissi' dance have been recognized using Kinect sensor, which is used for visual sensing. With the help of Kinect, we obtain a set of twenty body junction coordinates out of which only sixteen are required for our proposed work. A unique and simple methodology has been adopted to distinguish between the postures based on the distance and angle between the different joint coordinates. The average joint information values from each hand and leg are processed and they form the four vertices of a 4-sided polygon. The value of the four edges and four angles are to obtain from the polygon to train SVM. The experimental details show that this algorithm performs with a high recognition rate of 92.7% using SVM.