Dance Style Classification using Laban-Inspired and Frequency-Domain Motion Features

TL;DR

This work tackles dance style classification from motion data by proposing a lightweight, pose-based framework that combines Laban Movement Analysis–inspired temporal-spatial descriptors with FFT-based rhythmic features. Features are computed per frame relative to the hip center, encompassing Body, Shape, Space, and Effort components, plus velocity, acceleration, yaw, and spectral information, then aggregated over $N_s$ segments before being fed into lightweight classifiers. Across datasets (AIST, Motorica, ImperialDance), the approach achieves high accuracy, up to $99.00\%$, with strong generalization when using mixed training and segment-based majority voting. The method offers interpretable motion representations with low computational cost, suitable for education, dance analytics, and performance studies, and is complemented by comprehensive ablations and baselines. It demonstrates that well-structured, interpretable pose features can rival deeper models while providing transparency about the motion cues driving classification.

Abstract

Dance is an essential component of human culture and serves as a tool for conveying emotions and telling stories. Identifying and distinguishing dance genres based on motion data is a complex problem in human activity recognition, as many styles share similar poses, gestures, and temporal motion patterns. This work presents a lightweight framework for classifying dance styles that determines motion characteristics based on pose estimates extracted from videos. We propose temporal-spatial descriptors inspired by Laban Movement Analysis. These features capture local joint dynamics such as velocity, acceleration, and angular movement of the upper body, enabling a structured representation of spatial coordination. To further encode rhythmic and periodic aspects of movement, we integrate Fast Fourier Transform features that characterize movement patterns in the frequency domain. The proposed approach achieves robust classification of different dance styles with low computational effort, as complex model architectures are not required, and shows that interpretable motion representations can effectively capture stylistic nuances.

Figures (4)

• Figure 1: Sample images of different dance styles from the AIST dataset Tsuchida2019_aist.
• Figure 2: Schematic illustration of our dance style classification pipeline. In the first step, body keypoints are extracted from the input videos. In the next step, frame-wise features are extracted and then temporally related. Finally, these temporal features are divided into segments and fed into the classifiers.
• Figure 3: Confusion matrix for the different dance genres of the AIST dataset using gradient boosting and train/test split of mixed/mixed.
• Figure 4: Mean feature importance plot for the AIST dataset using the full feature set from 3D keypoints, gradient boosting as classifier and train/test split of mixed/mixed.