CGLER: Laban Effort Framework Analysis with Conducting Gestures Using Neural Networks

Gesture recognition has been a research area of considerable interest for many years. This is due to the many areas it can be applied to, from aiding with sign language recognition to allowing natural interfacing with a computer. With the increase of computing power, new methods of using computational intelligence (CI) algorithms have been explored. One way to improve CI is to test new frameworks for feature extraction and gesture description. One such framework is the Laban Movement Analysis (LMA) which provides functional and expressive descriptors for human gesture. In the past, LMA has been most commonly applied to theatre and dance. This paper attempts to apply LMA to the study of conducting gestures under a computational intelligence framework. In our method, deep learning was employed by using stacked neural networks models. The stacked neural network method allows for more layers of learning, which is anticipated to produce more accurate results. Three datasets of 72 total files was used to train five different neural networks. The results show that we have achieved high accuracy in classifying the various Laban qualities, yielding a maximum accuracy rate of 99.85% in one area. The main outcome of this research is to showing the use of deep learning in understanding and classifying Laban features and descriptors through musical conducting gestures.