University of California, Santa Barbara An Integrated System of 3D Motion Tracker and Spatialized Sound Synthesizer John Thompson (Music) Mary Li (ECE)

University of California, Santa Barbara

An Integrated System of 3D Motion Tracker and Spatialized

Sound Synthesizer

John Thompson (Music)

Mary Li (ECE)

Michael Quinn (ECE)

University of California, Santa Barbara

Goal

• Develop an interactive music synthesis system while exploring tracking and surveillance technologies, spatial music composition strategies, and sound synthesis techniques

University of California, Santa Barbara

Hardware & Software

• Unibrain Fire-I Cameras

• PC Running Windows XP

• Apple Powerbook and G5

• Intel’s OpenCV Libraries

• Max/MSP/Jitter

• SuperCollider

University of California, Santa Barbara

Project Summary

• 2D Tracking• Camera Calibration• 3D Position Calculation• Composition and Sound Synthesis

University of California, Santa Barbara

2D Tracking – Temporal Difference• Temporal Difference

– Subtract previous frame from the current frame to see what has changed.

University of California, Santa Barbara

2D Tracking - Background Subtraction• Develop a background model

• Subtract background from current frame

• Objects not in model will show up in the difference

University of California, Santa Barbara

2D Tracking - Thresholding

• Values chosen based on variance of background model

University of California, Santa Barbara

2D Tracking – Center of Mass

• For now, we assume that only one object is being tracked. Thus, the image center of mass approximates the object center of mass.

• Center of Mass is then sent to the 3D section.

University of California, Santa Barbara

Camera Calibration

• Purpose– A preparation for 3D estimation from 2D

images

• Methods– Matlab camera calibration toolbox– Intel OpenCV calibration functions

University of California, Santa Barbara

Camera Calibration -- intrinsic parameters

•Focal lengths: fx, fy

•Principal points: px, py

•Distortions: radial and tangential distortion coefficients

DirectShow Filter runs under MS Windows

University of California, Santa Barbara

Camera Calibration -- intrinsic parameters•Defines pixel coordinate points with respect to camera coordinate system

Ximage = Mintr Xcamera

Matlab Camera Calibration Toolbox

University of California, Santa Barbara

Camera Calibration -- extrinsic parameters•Defines camera coordinate points with respect to world coordinate system Xcamera = Mextr Xworld

OpenCV calibration routine (based on intrinsic parameters)

left camera view center camera view right camera view

University of California, Santa Barbara

3D Tracking -- methods

• Obtain 2D motion centroid information

•Epipolar Geometry

•Least Square

University of California, Santa Barbara

3D Tracking -- results

floor plan of visible space 18-pt tracking example

X_world Y_world Z_world

University of California, Santa Barbara

Tracking System Performance

•Realtime average

2.09 frames per second

• System performance can be improved by

1. Distributed computing: one PC for each camera

2. More cameras

3. Improve background segmentation

University of California, Santa Barbara

TransMedia SystemsTrans-media systems exist as independent engines behind artistic manifestations in diverse media.

Input:Tracking objects within a sensor space serves as a principle component to power the trans-media system.

Middle Layer:In the middle layer, the data from the Motion Tracking system is interpreted and labled. This data is then used to determine the activity and state of the sensor space.

Output:Specific media use the middle layer data to inform their processes. The result is projected into the sensor space. Interactivity is enhanced when the participants in the sensor space become aware of their relationship with the system.

Graphic Notations and Trans-media Systems: John Cage “Fontana Mix”

University of California, Santa Barbara

Spatial Composition Strategies -- Sonic Nodes

• A system of nodes are layed out in the virtual space. The system of nodes is comprised of Generative Nodes and Transformative Nodes

• The nodes have an activation space surrounding them. Tracked objects activate nodes at various levels depending on the tracked objects’ measured distance from the nodes’ center.

• The nodes adjust their positions over time to reflect the history of the space

University of California, Santa Barbara

When a tracked object moves within the activation space of a particular node, the node executes its action.

Figure 1

Different paths create unique realizations of phrase level material in the mobile form

Chord Nodes 34

Impulse Nodes 360

Sample Playback Nodes

50

Convolution Nodes

16

Pitch-Time Shift Nodes

64

University of California, Santa Barbara

Pitch sets in the chord nodes

[0 1 3 4][0 1 5 7][0 2 3 7][0 2 4 6][0 2 4 7][0 2 4 8][0 2 5 6][0 2 5 7][0 2 5 8][0 2 6 7][0 2 6 8][0 3 5 6][0 3 5 7][0 4 5 8]

Thirty-four chordNodes are scattered in the virtual space. Seventeen of the chordNodes contain a unique four note pitch set.

Fourteen of the Seventeen sets are unique in their normal order. Although the pitch sets are diverse, they are closely knitted in their makeup. This lends a unified quality to the pitched verticalities of the sonic space.

As multiple users move throughout the space, the sonic material shifts, melding the space into a cohesive flow.

The space adapts its pitched contents to the actions of the users of the space.

University of California, Santa Barbara

Sound Spatialization

The system outputs quadraphonic audio

distributed to speakers surrounding the sensor

space. The position of the sounds within the

sensor space is determined by the position of

the tracked object. (Figure 1)

Distance is simulated through direct sound

to reverberant sound mixture. This ratio is

dictated by the following formulas:

Direct Sound Amplitude1/zPosition.abs

Reverb Sound Amplitude = 1/zPosition.abs.sqrt

Figure 1

Sensor Space

Speakers

Object Position

University of California, Santa Barbara

• Nodes are represented by colored circles. When tracked objects move within their activation space, the object leaves a mark at that position.

• An object leaves traces of its paths and node activation.

Visualization

University of California, Santa Barbara

Future Work

• Improve the system by enabling the tracking of multiple objects as well as incorporating features such as shape, size, and color.

• Improve integration with the musical synthesis system.

University of California, Santa Barbara

• Professor B.S. Manjunath

• Professor G. Legrady

• Professor J. Kuchera-Morin

• Dr. Xinding Sun

• NSF IGERT Program

• Fellow IGERTers

Special Thanks

University of California, Santa Barbara

Q ??