Dance Posture/Steps Classification Using 3D Joints from the Kinect Sensors
Nikoloas Bakalos, Eftychios E. Protopapadakis, Anastasios D. Doulamis, Nikolaos D. Doulamis · 2018
In this paper, we introduce a new framework for find out the dance postures from dance sequences. The data exploited are from 3D point joints as being estimated by the Kinect-II sensor. The analysis, instead of other traditional methods, assigns the correct label to each frame of the Kinect-derived sequence with respect to the actual posture involved. On the contrary, the conventional dance classification methods categorize the dances to their styles, type of derived from tempo and music rhythm. Seven classifiers are tested to assess the performance in posture classification. The selected classifiers are either probabilistic or use linear or non-linear kernels such as the Support Vector Machines. We have also included ensemble classification schemes to improve the analysis. Experiments are given to real-world dances focusing on Greek folkloric ones. The results are objectively assessed under the precision and recall metrics and provide a comparative framework.