Dance Recognition System Using Lower Body Movement

Abstract

The current means of locating specific movements in film necessitate hours of viewing, making the task of conducting research into movement characteristics and patterns tedious and difficult. This is particularly problematic for the research and analysis of complex movement systems such as sports and dance. While some systems have been developed to manually annotate film, to date no automated way of identifying complex, full body movement exists. With pattern recognition technology and knowledge of joint locations, automatically describing filmed movement using computer software is possible. This study used various forms of lower body kinematic analysis to identify codified dance movements. We created an algorithm that compares an unknown move with a specified start and stop against known dance moves. Our recognition method consists of classification and template correlation using a database of model moves. This system was optimized to include nearly 90 dance and Tai Chi Chuan movements, producing accurate name identification in over 97% of trials. In addition, the program had the capability to provide a kinematic description of either matched or unmatched moves obtained from classification recognition.

Application of motion capture technology based on dance big data in dance retrieval

“…The motion capture system used in this paper is an optical motion capture system, which adopts the first optical capture instrument dims9100 in China. The main hardware equipment of the system includes: high-speed three-dimensional data dynamic capture system, dims controller, infrared sensor, and optical identification points [16]. Dancers are required to wear tight-fitting clothes and stick 29 reflective markers on key parts of their bodies.…”

The use of deep learning technology in dance movement generation

“…In practice, the OpenPose network first uses the first few network layers of Visual Geometry to extract features from images. Next, these features are transmitted to two parallel convolution layer branches (Simpson et al, 2014;Kim et al, 2021). The first branch is used to predict 18 confidence maps, each representing a joint in the human skeleton.…”

Theater as Data

“…Subsequently, they used this data to automatically identify, annotate, and retrieve film data (Wiesner 2012). Using biomechanical analysis, they were able to achieve high accuracy in the automatic classification of these movements (Simpson, Wiesner, and Bennett 2014). Acknowledging that dance is more complex than individual steps, they have also developed systems to identify conceptual metaphors in dance (such as "conflict").…”