Automatic Dancing Assessment Using Kinect

Abstract

This paper presents the design of an automatic assessing system for dance learners. By exploiting the Microsoft Kinect to acquire 3-D motion data of the learners, the proposed system extracts features and performs matching between the acquired data and the corresponding sample motion of a teacher. The proposed systems have several distinctive functions not appearing in other existing systems, including (1) the performance of a learner can be assessed separately in terms of accuracies in posture and rhythm and (2) the automatic identification of error motion on body articulations during the learn’s exercising. To tolerate the lagged, redundant and wrong motions, which are frequently seen on a naïve learn’s exercising, our assessment of posture accuracy is done by dynamic time warping (DTW) algorithm so that two motion sequences with different timing and lengths can be accurately aligned. On assessing the accuracy in rhythm, the system applies the fast Fourier transform (FFT) to extract the motion frequency which portrays the rhythm of body motions. Compared with the assessment made by human judges, the system’s assessment achieves the consistency of 84% and 77% in assessing the posture accuracy and tempo accuracy, respectively.