Emotion Recognition in Contemporary Dance Performances Using Laban Movement Analysis

Muhammad Turab, Philippe Colantoni, Damien Muselet, Alain Trémeau · 2025

Read Emotion Recognition in Contemporary Dance Performances Using Laban Movement Analysis by Muhammad Turab, Philippe Colantoni, Damien Muselet, Alain Trémeau (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-21154.txt.

Introduction

Muhammad Turab, Philippe Colantoni, Damien Muselet, and Alain Trémeau teach machines to read emotion from dance. Their 2025 paper uses Laban Movement Analysis, a system dancers use to describe movement quality, as the source of features. Dancers performed pieces under varied emotional states while their joint positions were captured.

The team improved existing Laban descriptors and added new ones that capture both measured and felt sides of movement. Classifiers such as random forests and support vector machines learned from those features. The best reached accuracy near ninety-seven percent at naming the emotion. Explainable methods then showed which features drove the calls, keeping the system readable rather than a black box. The authors point to uses in performance review, dance teaching, and human-computer interaction.

Catalog record

CREATORMuhammad Turab, Philippe Colantoni, Damien Muselet, Alain Trémeau
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE19 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-21154

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Emotion Recognition in Contemporary Dance Performances Using Laban Movement Analysis Muhammad Turab, Philippe Colantoni, Damien Muselet, Alain Trémeau Abstract This paper presents a novel framework for emotion recognition in contemporary dance by improving existing Laban Movement Analysis (LMA) feature descriptors and introducing robust, novel descriptors that capture both quantitative and qualitative aspects of the movement. Our approach extracts expressive characteristics from 3D keypoints data of professional dancers performing contemporary dance under various emotional states, and train multiple classifiers, including Random Forests and Support Vector Machines. Additionally, provide in-depth explanation of features and their impact on model predictions using explainable machine learning methods. Overall, our study improves emotion recognition in contemporary dance and offers promising applications in performance analysis, dance training, and human–computer interaction with highest accuracy of 96.85%. 1 Introduction In recent years, human emotion recognition has become an important research direction in the field of motion analysis and human-computer interaction. Emotion recognition involves the analysis of various emotional data by extracting features that describe the emotion and their relationship to classify the emotional state. It has been widely used in many fields including emotions in facial expressions [ 11 ] , speech recognition [ 16 ] , text [ 2 ] and psychological signals [ 15 ] . Emotion analysis is not limited to these fields, but it can expand to any field that involves movements. For instance, dance is a form of body movements that is used to express various emotions. Dance has been an integral part of human history, traditions, and cultures to express, tell stories, and convey emotions. Over the years, dance styles have evolved greatly reflecting cultural, social, and artistic changes. One of the new styles that has emerged is contemporary dance that combines elements of jazz, ballet, and modern dance. It originates from a desire to break away from the rigid structures of traditional ballet and modern dance. Unlike traditional dance, it allows dancers to explore novel forms of movement and emotional expression without setting strict rules. Due to the free-form structure of the contemporary dance style, it becomes challenging to recognize emotions. For such challenges, Laban Movement analysis (LMA) [ 17 ] provides a comprehensive framework for the evaluation and interpretation of human body movements. It is divided into four main components including Body, Effort, Shape, and Space. These components describe the structural, geometrical, and dynamic properties of motion. Body and Space components describe how the human body moves, either within the body or in relation with the 3D space surrounding the body. The Shape component describes the shape morphology of the body during the motion, whereas the Effort component focuses on the qualitative aspects of the movement in terms of dynamics, energy, and intent. LMA is more qualitative than quantitative, and a few studies have tried to propose feature descriptors that quantify the qualities of LMA to some extent. However, there are still a few challenges including capturing temporal dynamics of movement, and clear explanation of feature descriptors. Many current feature descriptors for movement analysis may not fully capture the dynamic characteristics without temporal dynamics. To address these challenges, we propose robust feature descriptors with temporal dynamics to capture the dynamic nuances of LMA qualities with their explanation on model predictions and individual performance. The main contributions of this study are as follows: • We improve existing feature descriptors by adding temporal dynamics to accurately capture movement trends and behaviors within short time windows. • We propose new feature descriptors for the four LMA components that capture rich representations of movements. • We perform a comprehensive evaluation on a contemporary dance dataset using Machine Learning (ML) methods, and analyze the interpretability of our descriptors with eXplainable AI methods. 2 Related Work Emotion recognition in dance movements has been an active field due to its applications in human-computer interaction [ 9 ] , performance analysis [ 24 ] , and evaluation [ 6 ] . Understanding how dancers express emotions through movement provides valuable insights into the expressive behavior of the performer. To extract expressive behavior from movement, LMA has become a popular framework as it provides a structured language to describe both the quantitative and qualitative aspects of human motion. In study [ 1 ] , LMA is used to recognize and analyze expressive gestures, including dancing, moving, waving, pointing, and stopping, with four emotions happy, angry, sad and neutral. Feature evaluation was performed using ML methods and subjective evaluation. The results show that the features are 87. 03% accurate in gesture classification. [ 25 ] used Kinect to track dancers and developed models for action and emotion recognition with 88.34% and 98.95% accuracy on skeleton data. Features like eigenvalue speed and skeleton pair distance help distinguish emotions. LMA was used in human action recognition in [ 22 ] , with a two-stage approach in which the first feature descriptors are defined using three components of LMA, including body, space, and shape. The results were obtained on various datasets using Dynamic Time Warping (DTW) to measure similarity in actions. Several studies have focused on the classification of dance emotions using LMA-based feature descriptors. A method is proposed to extract the motion qualities from dance performances for indexing and analysis in [ 4 ] . LMA-based features were used to investigate the correlation of features with dancers expressed emotional state. The results show low inter-feature correlation between dancers and their emotional state. [ 3 ] investigated similarities of various emotional states using extracted LMA extracted features. The results show that the features can partially extract LMA components that can be used for the dance classification from emotion. For automatic feature extraction and better results, Deep learning (DL) has been used. [ 27 ] applied a hybrid deep learning method based on Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) to effectively identify emotions. For body structure, spatial orientation, and force effect LMA-based feature descriptors were used to add emotional changes during movement. Their approach successfully recognizes emotions in dance movements with high accuracy. 3 Methodology To improve emotion recognition in contemporary dance, our study is divided into six main parts: 1) Data pre-processing, 2) 3D human body pose estimation, 3) LMA feature extraction, 4) Multiclass classification, 5) Evaluation, and 6) Explainability of feature descriptors as shown in Fig. 1 . Figure 1 : Overview of the proposed method for Emotion Recognition in Contemporary Dance Performances. 3.1 Data Pre-processing We use contemporary dance performance videos from the Dance Motion capture dataset provided by the University of Cyprus [ 5 ] . This dataset contains performances of 5 dancers expressing 12 emotions including afraid, angry, annoyed, bored, excited, happy, miserable, pleased, relaxed, sad, satisfied and tired. The dataset has a few limitations, such as: 1) inconsistent video duration, 2) limited number of dancers and performances, 3) low frame rate, and 4) a fixed camera angle for a single perspective. 3.2 3D Human Body Pose Estimation To obtain the joints keypoint data from dance videos, we use Neural Localizer Fields (NLF) [ 23 ] for 3D body pose estimation. We compare NLF with other state-of-the-art (SOTA) methods including Mediapipe [ 19 ] and OpenPose [ 12 ] and choose NLF for several reasons including: 1)

Read the complete text