DanceIR

DanceIR

A Research Framework for Dance Information Retrieval and Motion-Based Tempo Estimation

DanceIR is a research framework and Python toolbox for analyzing rhythmic structure directly from dance movement, without relying on audio. It combines principles from Music Information Retrieval (MIR), signal processing, and movement science to extract rhythmic anchors, estimate dance-derived tempo, and study body–music synchrony using motion capture and pose data.

This page provides an overview of the research, method, and supplementary demonstrations accompanying the DanceIR toolbox and related publications.

Overview

Natural human movement contains strong rhythmic structure. Dancers often articulate beats, subdivisions, and metrical layers through coordinated hand, foot, and torso motions. DanceIR formalizes these patterns as rhythmic anchors—kinematic events such as:

  • trajectory endpoints (zero-velocity points),
  • velocity peaks,
  • motion-energy peaks.

These anchors form periodic sequences that reveal the dancer’s internal pulse. DanceIR tracks these sequences over time and derives a global tempo estimate aligned with the dancer’s movement patterns.

The framework is designed for research on dance–music interaction, multimodal rhythm analysis, and embodied entrainment.

Method Summary

1. Rhythmic Anchor Extraction

DanceIR identifies movement anchors from pose or motion-capture data.
Common anchor types include:

  • Zero-velocity anchors (trajectory reversals)
  • Peak-velocity anchors
  • Motion-energy peaks
  • Foot contact events (from foot-ground interaction)

These sequences represent the dancer’s primary periodic motions.

2. Periodicity & Tempo Estimation

Tempo is estimated by analyzing anchor sequences using:

  • inter-onset interval (IOI) statistics,
  • windowed periodicity analysis,
  • log-scaled tempo tracking,
  • a novel Octave-Invariant Accuracy metric.

DanceIR outputs both:

  • continuous tempo curves, and
  • global tempo estimates per performance.

3. Octave-Invariant Metric (ΓOI)

To quantify dance-derived tempo accuracy, DanceIR introduces the continuous, log-scaled Octave-Invariant Accuracy:

  • Perfect alignment at half-time, original tempo, or double-time → ΓOI = 1
  • Outside the tolerance band → ΓOI = 0
  • A tempo estimate is considered a hit when ΓOI > 0

This provides a musically meaningful evaluation consistent with metrical hierarchy.

Supplementary Videos

Below are selected demonstrations generated using DanceIR.
Videos illustrate anchor extraction, periodicity tracking, global tempo estimation, and cross-anchor comparisons.

Note: Videos hosted in this repository:
https://sagar0dutta.github.io/danceir-supplementary/videos/

Anchor Extraction Example

Identification of zero- and peak-velocity anchors from hand trajectories.

Global Tempo Estimation (ΓOI)

Tempo tracking using rhythmic anchors and octave-invariant scoring.

Comparison Across Anchor Types

Different body segments emphasize different metrical layers; DanceIR unifies them.

DanceIR Toolbox

DanceIR is available as a Python package:

```bash pip install danceir