1 of 39

MUSIC-DRIVEN SYSTEM FOR GROUP DANCE FORMATION GENERATION

​

25-26J-358

“SyncForm”

2 of 39

25-26J-358

09/01/2026

01

MEMBERS

Supervisor

Ms. Jenny Krishara

Co .Supervisor

Ms. Dinuka Wijendra

External Supervisor

Mr. Shakitha Kanchana

Shaki Dance Studio

External Supervisor

Ms. Sachini (Fiya)

The Beez - M Entertaintment(Sirasa TV)

3 of 39

25-26J-358

09/01/2026

02

MEMBERS

Team Leader

Dissanayake D.M.S.A.B

IT22595294

Uthpalani A.K.M.

IT22327444

Attanayake J.S

IT22911230

Samarakoon K.K

IT22346940

4 of 39

25-26J-358

09/01/2026

03

INTRODUCTION

5 of 39

Research Problem & Impact

09/01/2026

04

    • Manual formation planning is slow and requires many iterations.

​

    • Beginner dancers cannot create effective formations on their own, and a professional choreographer is often required.

​

    • Hard to align group formations to beats/phrases while avoiding dancer collisions.

​

    • Existing tools are manual; research mostly targets body motion, not formation-level planning.

​

    • Impact: waste of time, as a lot of time has to be spent during these processes

​

6 of 39

Proposed Solution

09/01/2026

05

    • AI-assisted, music-driven formation generation for group dance.

​

    • Beat/phrase-aware formation layouts (Diffusion model on music embeddings).

​

    • Collision-free, beat-timed transitions between formations (assignment +agent planning).

​

    • Formation evaluation (FQI + classifier) + Style control with an interactive web app.

​

7 of 39

09/01/2026

06

Methodology

    • Audio → Feature Extraction & Segmentation → Formation Generator → Transition Planner → Evaluation

​

    • React Web App for visualization, editing, playback, and export.

​

    • Integration via API endpoints for model inference.

​

    • Frontend: formation canvas, controls, timeline & playback.

​

    • Backend: audio embeddings/segments, formation generation, transition planning, evaluation.

​

    • Demo: upload music → detect beats → generate formations → generate transitions → preview → export.

​

    • Two key functionalities + integration will be demonstrated.

​

8 of 39

09/01/2026

07

Overall System Diagram

9 of 39

IT22327444

Uthpalani A.K.M.

25-26J-358

09/01/2026

08

A.K.M. UTHPALANI

IT22327444

BSC (HONS) DEGREE IN INFORMATION TECHNOLOGY (SPECIALIZATION IN INFORMATION TECHNOLOGY)

​

MUSIC FEATURE EXTRACTION & SEGMENTATION

10 of 39

MUSIC FEATURE EXTRACTION & SEGMENTATION

IT22327444

Uthpalani A.K.M.

25-26J-358

09/01/2026

09

COMPONENT SCOPE & PROBLEM DEFINITION

Sub-Problem

​

Formations must align with beats, downbeats, and phrases; manual timing is error-prone and inconsistent across songs.

​

Solution

​

Extract rhythm-aware embeddings + segment boundaries to provide beat/phrase timeline for formation generation and transition timing.

​

​

11 of 39

MUSIC FEATURE EXTRACTION & SEGMENTATION

IT22327444

Uthpalani A.K.M.

25-26J-358

09/01/2026

10

FEATURE COMPARISON

12 of 39

MUSIC FEATURE EXTRACTION & SEGMENTATION

IT22327444

Uthpalani A.K.M.

25-26J-358

09/01/2026

10

FEATURE COMPARISON

13 of 39

MUSIC FEATURE EXTRACTION & SEGMENTATION

IT22327444

Uthpalani A.K.M.

25-26J-358

09/01/2026

11

Audio

How it Works

audio analyzing (librosa)

beat/downbeat + phrase segmentation(intro/verse/Chorus)

Formation points timestamps + segment features.

14 of 39

MUSIC FEATURE EXTRACTION & SEGMENTATION

IT22327444

Uthpalani A.K.M.

25-26J-358

09/01/2026

12

DESIGN EXcellence/Contribution

15 of 39

MUSIC FEATURE EXTRACTION & SEGMENTATION

IT22327444

Uthpalani A.K.M.

25-26J-358

09/01/2026

13

COMPONENT DIAGRAM

16 of 39

IT22911230

Attanayake J.S

25-26J-358

09/01/2026

14

J.S. ATTANAYAKE

IT22911230

BSC (HONS) DEGREE IN INFORMATION TECHNOLOGY (SPECIALIZATION IN INFORMATION TECHNOLOGY)

FORMATION LAYOUT PLANNING

17 of 39

IT22911230

Attanayake J.S

25-26J-358

09/01/2026

15

FORMATION LAYOUT PLANNING

COMPONENT SCOPE & PROBLEM DEFINITION

Sub-Problem

​

Existing systems generate motions or allow manual formations; they do not generate group layouts conditioned on music structure.

Solution

​

Generate multiple feasible 2D formations (x,y positions) conditioned on music segment embeddings and dancer count.

​

18 of 39

Input: dancer count + music embedding and segments

How it Works

Diffusion model

sample diverse layouts

Apply constraints (stage bounds, min distance)

IT22911230

Attanayake J.S

25-26J-358

09/01/2026

16

FORMATION LAYOUT PLANNING

19 of 39

FEATURE COMPARISON

IT22911230

Attanayake J.S

25-26J-358

09/01/2026

17

FORMATION LAYOUT PLANNING

SyncForm

20 of 39

IT22911230

Attanayake J.S

25-26J-358

09/01/2026

18

FORMATION LAYOUT PLANNING

DESIGN EXcellence/Contribution

21 of 39

IT22911230

Attanayake J.S

25-26J-358

09/01/2026

19

FORMATION LAYOUT PLANNING

COMPONENT DIAGRAM

22 of 39

D.M.S.A.B. DISSANAYAKE

IT22595294

BSC (HONS) DEGREE IN INFORMATION TECHNOLOGY (SPECIALIZATION IN INFORMATION TECHNOLOGY)

IT22595294

Dissanayake D.M.S.A.B

25-26J-358

09/01/2026

20

FORMATION TRANSITION PLANNING & INTERPOLATION

23 of 39

FORMATION TRANSITION PLANNING & INTERPOLATION

Sub-Problem

​

Even with good start formations and end formations, transitions

Collision Risk: Without guidance, dancers moving to new formations (e.g., Line to Circle) may cross paths inefficiently can collide or miss beats.

Solution

​

Assign dancers to target positions and compute collision-free, beat-timed paths with smooth interpolation.

COMPONENT SCOPE & PROBLEM DEFINITION

IT22595294

Dissanayake D.M.S.A.B

25-26J-358

09/01/2026

21

24 of 39

FORMATION TRANSITION PLANNING & INTERPOLATION

FEATURE COMPARISON

SyncForm

IT22595294

Dissanayake D.M.S.A.B

25-26J-358

09/01/2026

22

25 of 39

How it Works

IT22595294

Dissanayake D.M.S.A.B

25-26J-358

09/01/2026

23

FORMATION TRANSITION PLANNING & INTERPOLATION

1. Transition Orchestration

    • The system receives the "Start" and "End" formation positions from the pipeline.
    • It automatically generates intermediate frames to fill the time gap between formations.

​

2.Optimal Assignment Logic

    • Implements the Hungarian Algorithm.
    • Calculates the distance from every current dancer position to every target spot.
    • Solves for the pairing that results in the shortest total distance, preventing unnecessary stage crossing.

​

3.Motion Synthesis (RL Agent)

    • Uses a Proximal Policy Optimization (PPO) neural network to determine dancer velocity and direction.
    • Inputs 41 state variables (current position, target position, neighbors dancers).
    • Outputs a smooth, collision free music sync timeline

​

26 of 39

FORMATION TRANSITION PLANNING & INTERPOLATION

DESIGN EXcellence/Contribution

IT22595294

Dissanayake D.M.S.A.B

25-26J-358

09/01/2026

24

27 of 39

IT22595294

Dissanayake D.M.S.A.B

25-26J-358

09/01/2026

25

FORMATION TRANSITION PLANNING & INTERPOLATION

COMPONENT DIAGRAM

28 of 39

K.K. SAMARAKOON

IT22346940

BSC (HONS) DEGREE IN INFORMATION TECHNOLOGY (SPECIALIZATION IN INFORMATION TECHNOLOGY)

IT22346940

Samarakoon K.K

25-26J-358

09/01/2026

26

FORMATION STYLE CONTROL & EVALUATION

29 of 39

FORMATION STYLE CONTROL & EVALUATION

IT22346940

Samarakoon K.K

25-26J-358

09/01/2026

27

Sub-Problem

​

Formation quality and style are judged subjectively, no standardized way to score symmetry/spacing or genre alignment.

​

Solution

​

    • Formation Quality Index (FQI)
    • ML style classifier to score technical quality and stylistic fit.

​

COMPONENT SCOPE & PROBLEM DEFINITION

30 of 39

FEATURE COMPARISON

SyncForm

IT22346940

Samarakoon K.K

25-26J-358

09/01/2026

28

FORMATION STYLE CONTROL & EVALUATION

31 of 39

IT22346940

Samarakoon K.K

25-26J-358

09/01/2026

29

How it Works

1. Formation Quality Index (FQI)

2. Style Classification Engine (AI)

​

    • Converts raw (X,Y) coordinates into Gaussian Heatmaps (Density Images).
    • Uses a CNN to automatically recognize shapes.
    • This transformation allows the AI to "see" visual topology just like a human eye does.

3. Physical Feasibility

    • Ensures that generated movements are physically possible for a human.
    • Calculates frame to frame Velocity and Acceleration for every single dancer to detect impossible movements.
    • Evaluates five core metrics (Symmetry, Balance, Compactness, Spacing, and Center Bias)
    • A mathematical engine designed to measure the objective quality of a dance formation.

FORMATION STYLE CONTROL & EVALUATION

32 of 39

IT22346940

Samarakoon K.K

25-26J-358

09/01/2026

30

DESIGN EXcellence/Contribution

FORMATION STYLE CONTROL & EVALUATION

33 of 39

IT22346940

Samarakoon K.K

25-26J-358

09/01/2026

31

COMPONENT DIAGRAM

FORMATION STYLE CONTROL & EVALUATION

34 of 39

09/01/2026

32

Risk Mitigation

Risk

Mitigation Strategy

Music analysis inaccuracies

Robust audio preprocessing with automated checks and human validation

Unsafe or impractical formations

Enforce spacing, stage boundary, and collision constraints

Poor transition smoothness

Beat-aligned timing and smooth, learned movement paths

Style or aesthetic mismatch

Style control parameters with automated and user-based evaluation

35 of 39

09/01/2026

33

Commercialization

Pricing Model

Target Market

Value Proposition

    • First Music - Driven Formation generator
    • collaborating with an external supervisor to build a strong brand identity for the product.
    • Free: Community Dancers
    • $20/mo : Schools / Small Teams
    • $50/mo : Pro Studios
    • Dance Schools & studios
    • Cover Groups, Uni clubs, choreographers
    • Music - driven Automation
    • Style - Adaptive

(hip - hop now)

    • Built in Formation Quality Index (FQI)

COMMERCIALIZATION & BUSINESS POTENTIAL

36 of 39

09/01/2026

34

User Feedback on Prototype

External Supervisor

Ms. Sachini (Fiya)

The Beez - M Entertainment(Sirasa TV)

37 of 39

09/01/2026

35

Appendix

38 of 39

09/01/2026

39 of 39

THANK YOU!

09/01/2026