Kinect Tutorial

Kinect Tutorial
Nicholas Gillian
Responsive Environments, MIT Media Lab
Thursday, September 12th, 2013
Thursday, September 12, 13

Kinect
RGB Camera
3D Depth Sensors
•Depth map is constructed by analyzing a speckle pattern
of infrared laser light
•Body position is inferred using machine learning
•Kinect uses structured light

Kinect
•Structured light: project a known pattern onto the scene
and infer depth from the deformation of that pattern
Jason Geng, Structured-light 3D surface imaging: a tutorial,
Advances in Optics and Photonics,Vol. 3, Issue 2, pp. 128-160 (2011)
Zhang et al, 3DPVT (2002)

Kinect
•Kinect uses an astigmatic lens with different focal length
in x- and y directions
•The lens causes a projected circle to become an ellipse
whose orientation depends on depth

Kinect
•The Kinect also uses parallax, i.e. if you look at a scene
from different angle, things that are closer get shifted to
the side more than things that are far away
•The Kinect analyzes the shift of the speckle pattern by
projecting from one location and observing from another
3D Depth Sensors

Kinect
Jamie Shutton et. al, Real-Time Human Pose Recognition in Parts from Single Depth Images, CVPR, 2011

Kinect
100K poses 1 million training samples

Kinect
•Randomized Decision Forests
The probability of pixel u
belonging to body part c is:

Kinect
•Features

Kinect Libraries, APIs & Tools
•OpenNI SDK
•Openframeworks
•Microsoft Ofﬁcial Kinect SDK
•Synapse

OpenNI - Installation
•OpenNI
•NITE
•Sensor Kinect

OpenNI - Data
RGB Image Depth Image
Label Image
[640 480]
Depth: Unsigned short
Label: Unsigned short
RGB: Unsigned char

OpenNI - Generators
XnGenerator
AudioGenerator GestureGenerator HandsGenerator UserGenerator ...
//For example
DepthGenerator depthGenerator;
ImageGenerator imageGenerator;

OpenNI - Configuration
<OpenNI>

<Licenses>

<License vendor="PrimeSense" key="0KOIk2JeIBYClPWVnMoRKn5cdY4="/>

</Licenses>

<Log writeToConsole="true" writeToFile="false">



<LogLevel value="3"/>

<Masks>

<Mask name="ALL" on="false"/>

</Masks>

<Dumps>

</Dumps>

</Log>

<ProductionNodes>

<Node type="Depth">

<Configuration>

<Mirror on="true"/>

</Configuration>

</Node>

<Node type="Image" stopOnError="false" />

<Node type="User" />

</ProductionNodes>
</OpenNI>
This key is common
PrimeSense key
Enable the depth camera
Enable the RGB camera
Enable the user tracker

OpenNI - Manual Conﬁguration
XnStatus nRetVal = context.FindExistingNode(XN_NODE_TYPE_DEPTH, depthGenerator);

OpenNI - Manual Conﬁguration
XnStatus nRetVal = context.FindExistingNode(XN_NODE_TYPE_DEPTH, depthGenerator);
if (nRetVal != XN_STATUS_OK){
xn::MockDepthGenerator mockDepth;
nRetVal = mockDepth.Create(context);
// set some defaults
XnMapOutputMode defaultMode;
defaultMode.nXRes = 640;
defaultMode.nYRes = 480;
defaultMode.nFPS = 30;
nRetVal = mockDepth.SetMapOutputMode(defaultMode);
// set FOV
XnFieldOfView fov;
fov.fHFOV = 1.0225999419141749;
fov.fVFOV = 0.79661567681716894;
nRetVal = mockDepth.SetGeneralProperty(XN_PROP_FIELD_OF_VIEW, sizeof(fov), &fov);
XnUInt32 nDataSize = defaultMode.nXRes * defaultMode.nYRes * sizeof(XnDepthPixel);
XnDepthPixel* pData = (XnDepthPixel*)xnOSCallocAligned(nDataSize, 1, XN_DEFAULT_MEM_ALIGN);
nRetVal = mockDepth.SetData(1, 0, nDataSize, pData);
CHECK_RC(nRetVal, "set empty depth map");
depthGenerator = mockDepth;
}

OpenNI - Accessing Data
// Read next available data
context.WaitOneUpdateAll( userGenerator );
userSelector->UpdateFrame();
//Variables
UserGenerator userGenerator;
Context context;

OpenNI - Get the depth data
//Get the latest depth and image meta data
depthGenerator.GetMetaData( depthMD );
//Get a pointer to the depth data
const XnDepthPixel* pDepth = depthMD.Data();
//Loop over the depth pixels
for (XnUInt y = 0; y < depthMD.YRes(); y++){
for (XnUInt x = 0; x < depthMD.XRes(); x++){
//Access the current depth value
*pDepth;
//Increase the depth pixel
pDepth++;
}
}
//Variables
DepthGenerator depthGenerator;
DepthMetaData depthMD;

OpenNI - Get the RGB data
//Get the latest image meta data
imageGenerator.GetMetaData( imageMD );
//Get a pointer to the image data
const XnRGB24Pixel* pImage = imageMD.RGB24Data();
//Loop over the image pixels
for (XnUInt y = 0; y < imageMD.YRes(); y++){
for (XnUInt x = 0; x < imageMD.XRes(); x++){
//Access the current pixel value
* pImage;
//Increase the pixel pointer
pImage++;
}
}
//Variables
ImageGenerator imageGenerator;
ImageMetaData imageMD;

OpenNI - Skeleton data
//Variables
UserGenerator userGenerator;
XnUserID userIDs[ numTrackedUsers ];
XnUInt16 numUsers = userGenerator->GetNumberOfUsers();
userGenerator->GetUsers(userIDs, numUsers);
for( int i=0; i < numUsers; i++ ){
if( userGenerator->GetSkeletonCap().IsTracking( userIDs[i] ) ){
XnPoint3D userCenterOfMass;
userGenerator->GetSkeletonCap().IsCalibrating( userIDs[i] );
userGenerator->GetSkeletonCap().IsCalibrated( userIDs[i] );
userGenerator->GetCoM( userIDs[i], userCenterOfMass );
XnSkeletonJointTransformation jointData;
if( userGenerator->GetSkeletonCap().IsJointAvailable( XN_SKEL_HEAD ) ){
userGenerator->GetSkeletonCap().GetSkeletonJoint(userIDs[i], XN_SKEL_HEAD, jointData );
}
}
}

Kinect Gesture Recognition

SUPERVISED LEARNING

Supervised Learning
Training Data

Supervised Learning
Input Vector
Training Data

Supervised Learning
Input Vector
Training Data
{Feature Vector}

Supervised Learning
Input Vector
Training Data
{Feature Vector}
Feature

Supervised Learning
Input Vector
Training Data
{Feature Vector}
Feature
{Attribute}

Supervised Learning
Input Vector Target Vector
Training Data

Supervised Learning
Learning Algorithm
Training Data

Supervised Learning
Learning Algorithm
Model
Training Data

Supervised Learning
Learning Algorithm
Prediction
New Datum
Model
Training Data

Supervised Learning
Learning Algorithm
Prediction
Predicted Class
Class A
Model
Training Data
New Datum

Kinect Gesture Recognition
Synapse
OSC
Skeleton Data
Openframeworks
www.nickgillian.com/wiki/pmwiki.php/GRT/
OpenframeworksKinectExample

Kinect Code Tutorial Slides
www.nickgillian.com/09-12-13.html

Kinect Tutorial

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to Kinect Tutorial (20)

More from Camera Culture Group, MIT Media Lab (13)

Recently uploaded (20)

Kinect Tutorial