frog
Class GestureHMM

java.lang.Object
  frog.GestureHMM

All Implemented Interfaces:

GestureModel, Serializable

public class GestureHMM
extends Object
implements GestureModel, Serializable

GestureHMM is a modeling class that represents a gesture. As a model, it requires data (gesture instances) as well as parameters for training. GestureHMM utilizes a statistical HMM (as its name suggests) model as well as a Baum-Welch training method.

Author:

Team Better Recognize

See Also:

Serialized Form

Field Summary

Fields inherited from interface frog.GestureModel

GESTUREHMM, NUM_TYPES_SUPPORTED

Constructor Summary

GestureHMM(String n)
Constructs a GestureHMM with a name and default training parameters.

GestureHMM(String n, byte[] iconBuf)
Constructs a GestureHMM with a name and image.

GestureHMM(String n, GestureHMMParameters gParams)
Constructs a GestureHMM with a name and training parameters.

GestureHMM(String n, GestureHMMParameters gParams, byte[] iconBuf)
Constructs a GestureHMM with a name, training parameters, and an image.

GestureHMM(String n, GestureHMMParameters gParams, KmeansParameters kParams)
Constructs a GestureHMM with a name, training parameters, and quantizer parameters (Kmeans in particular).

GestureHMM(String n, GestureHMMParameters gParams, KmeansParameters kParams, byte[] iconBuf)
Constructs a GestureHMM with a name, training parameters, a gesture set, and an image.

GestureHMM(String n, GestureHMMParameters gParams, KmeansParameters kParams, Vector<Vector<Accel3D>> instances)
Constructs a GestureHMM with a name, training parameters, quantizer parameters (Kmeans in particular), and a training set.

GestureHMM(String n, GestureHMMParameters gParams, KmeansParameters kParams, Vector<Vector<Accel3D>> instances, byte[] iconBuf)
Constructs a GestureHMM with a name, training parameters, a gesture set, and an image.

GestureHMM(String n, GestureHMMParameters gParams, Vector<Vector<Accel3D>> instances)
Constructs a GestureHMM with a name, training parameters, as well as a gesture set.

GestureHMM(String n, GestureHMMParameters gParams, Vector<Vector<Accel3D>> instances, byte[] iconBuf)
Constructs a GestureHMM with a name, training parameters, a gesture set, and an image.

GestureHMM(String n, Vector<Vector<Accel3D>> instances)
Constructs a GestureHMM with a name and gesture set as well as default training parameters.

Method Summary

boolean	addAccel(int inum, Accel3D acc) Adds an additional Accel3D to the end of a gesture instance.
void	addInstance(Vector<Accel3D> instance) Adds an entire gesture instance to the gesture set.
void	clearStats() Clears all saved statistics associated with this GestureModel
void	correct() Signifies a correct recognition event to the GestureModel.
double	getAverageRecognitionProbability() Returns the average certainty currently associated with this GestureModel.
double	getDefaultProbability() For GestureHMM, default probability is obtained by averaging the probabilities with which each instance in the set used to create this model match the model.
BufferedImage	getIcon() Returns the icon of this gesture as a BufferedImage.
String	getName() Returns the name associated with this GestureModel.
int	getNumCorrect() Accessor for the number correct statistic.
int	getNumIncorrect() Accessor for the number incorrect statistic.
int	getNumNotRecognized() Accessor for the number not recognized statistic.
int	getNumStates() Returns the number of hidden states in this GestureHMM.
double	getProbability(Vector<Accel3D> instance) For GestureHMM, probability is computed using the forward algorithm.
Vector<Vector<Accel3D>>	getTrainingSet() Accessor for the training set of gesture instances associated with this GestureModel
double	getTrainingThreshold() Returns the training threshold of this GestureHMM.
int	getType() Accessor for the GestureModel type index.
void	incorrect() Signifies an incorrect recognition event to the GestureModel.
void	matchedWithProbability(double prob) Signifies a recognition event that occurred with a given probability.
void	notRecognized() Signifies an unrecognized recognition event to the GestureModel.
boolean	removeAccel(int inum, int anum) Removes an acceleration value from the gesture set.
boolean	removeInstance(int inum) Removes a gesture instance from the gesture set.
void	setIcon(byte[] imageBuffer) Associates a byte array image with the GestureModel
void	setName(String n) Associates a representative name with this GestureModel
boolean	train() Trains the GestureModel, constructing everything necessary for recognition to take place.
boolean	train(GestureHMMParameters gParams) Trains the GestureHMM with a specified set of training parameters.
boolean	train(GestureHMMParameters gParams, KmeansParameters kParams) Trains the GestureHMM with a specified set of training parameters as well as a specified set of quantizer parameters.
boolean	train(KmeansParameters params) Trains the GestureHMM with a specified set of quantizer parameters.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

GestureHMM

public GestureHMM(String n)

Constructs a GestureHMM with a name and default training parameters.

Parameters:

n - a name to associate with the GestureHMM

GestureHMM

public GestureHMM(String n,
                  Vector<Vector<Accel3D>> instances)

Constructs a GestureHMM with a name and gesture set as well as default training parameters.

Parameters:

n - a name to associate with the GestureHMM

instances - the gesture set for training the GestureHMM

GestureHMM

public GestureHMM(String n,
                  GestureHMMParameters gParams)

Constructs a GestureHMM with a name and training parameters.

Parameters:

n - a name to associate with the GestureHMM

gParams - HMM parameters for training

GestureHMM

public GestureHMM(String n,
                  GestureHMMParameters gParams,
                  Vector<Vector<Accel3D>> instances)

Constructs a GestureHMM with a name, training parameters, as well as a gesture set.

Parameters:

n - a name to associate with the GestureHMM

gParams - HMM parameters for training

instances - the gesture set for training the GestureHMM

GestureHMM

public GestureHMM(String n,
                  GestureHMMParameters gParams,
                  KmeansParameters kParams)

Constructs a GestureHMM with a name, training parameters, and quantizer parameters (Kmeans in particular).

Parameters:

n - a name to associate with the GestureHMM

gParams - HMM parameters for training

kParams - Kmeans parameters for training the quantizer

GestureHMM

public GestureHMM(String n,
                  GestureHMMParameters gParams,
                  KmeansParameters kParams,
                  Vector<Vector<Accel3D>> instances)

Constructs a GestureHMM with a name, training parameters, quantizer parameters (Kmeans in particular), and a training set.

Parameters:

n - a name to associate with the GestureHMM

gParams - HMM parameters for training

kParams - Kmeans parameters for training the quantizer

instances - the gesture set for training the GestureHMM

GestureHMM

public GestureHMM(String n,
                  byte[] iconBuf)
           throws IOException

Constructs a GestureHMM with a name and image.

Parameters:

n - a name to associate with the GestureHMM

iconBuf - the byte array image to associate with the GestureHMM

Throws:

IOException - thrown if image reading fails

GestureHMM

public GestureHMM(String n,
                  GestureHMMParameters gParams,
                  byte[] iconBuf)
           throws IOException

Constructs a GestureHMM with a name, training parameters, and an image.

Parameters:

n - a name to associate with the GestureHMM

gParams - HMM parameters for training

iconBuf - the byte array image to associate with the GestureHMM

Throws:

IOException - thrown if image reading fails

GestureHMM

public GestureHMM(String n,
                  GestureHMMParameters gParams,
                  Vector<Vector<Accel3D>> instances,
                  byte[] iconBuf)
           throws IOException

Constructs a GestureHMM with a name, training parameters, a gesture set, and an image.

Parameters:

n - a name to associate with the GestureHMM

gParams - HMM parameters for training

instances - the gesture set for training the GestureHMM

iconBuf - the byte array image to associate with the GestureHMM

Throws:

IOException - thrown if image reading fails

GestureHMM

public GestureHMM(String n,
                  GestureHMMParameters gParams,
                  KmeansParameters kParams,
                  byte[] iconBuf)
           throws IOException

Constructs a GestureHMM with a name, training parameters, a gesture set, and an image.

Parameters:

n - a name to associate with the GestureHMM

gParams - HMM parameters for training

kParams - Kmeans parameters for training the quantizer

iconBuf - the byte array image to associate with the GestureHMM

Throws:

IOException - thrown if image reading fails

GestureHMM

public GestureHMM(String n,
                  GestureHMMParameters gParams,
                  KmeansParameters kParams,
                  Vector<Vector<Accel3D>> instances,
                  byte[] iconBuf)
           throws IOException

Constructs a GestureHMM with a name, training parameters, a gesture set, and an image.

Parameters:

n - a name to associate with the GestureHMM

gParams - HMM parameters for training

kParams - Kmeans parameters for training the quantizer

instances - the gesture set for training the GestureHMM

iconBuf - the byte array image to associate with the GestureHMM

Throws:

IOException - thrown if image reading fails

Method Detail

addAccel

public boolean addAccel(int inum,
                        Accel3D acc)

Description copied from interface: GestureModel

Adds an additional Accel3D to the end of a gesture instance.

Specified by:

addAccel in interface GestureModel

Parameters:

inum - The index of the gesture instance to modify

acc - The Accel3D to insert

Returns:

a boolean signifying success or failure of the insertion

addInstance

public void addInstance(Vector<Accel3D> instance)

Description copied from interface: GestureModel

Adds an entire gesture instance to the gesture set.

Specified by:

addInstance in interface GestureModel

Parameters:

instance - The instance to insert

clearStats

public void clearStats()

Description copied from interface: GestureModel

Clears all saved statistics associated with this GestureModel

Specified by:

clearStats in interface GestureModel

correct

public void correct()

Description copied from interface: GestureModel

Signifies a correct recognition event to the GestureModel. Statistics for the model are updated accordingly.

Specified by:

correct in interface GestureModel

getDefaultProbability

public double getDefaultProbability()

For GestureHMM, default probability is obtained by averaging the probabilities with which each instance in the set used to create this model match the model. In other words, it average the value of getProbability obtained with each gesture instance in the training set.

Specified by:

getDefaultProbability in interface GestureModel

Returns:

the default probability of this model.

getIcon

public BufferedImage getIcon()

Description copied from interface: GestureModel

Returns the icon of this gesture as a BufferedImage.

Specified by:

getIcon in interface GestureModel

Returns:

the icon of this gesture; or null if the gesture has no icon.

getName

public String getName()

Description copied from interface: GestureModel

Returns the name associated with this GestureModel. (e.g. a model for a circle might appropriately be named "Circle")

Specified by:

getName in interface GestureModel

Returns:

the name of the GestureModel.

getNumCorrect

public int getNumCorrect()

Description copied from interface: GestureModel

Accessor for the number correct statistic. This value is entirely dependent on calls to GestureModel.correct()

Specified by:

getNumCorrect in interface GestureModel

Returns:

the saved number of correct recognition events.

getNumIncorrect

public int getNumIncorrect()

Description copied from interface: GestureModel

Accessor for the number incorrect statistic. This value is entirely dependent on calls to GestureModel.incorrect()

Specified by:

getNumIncorrect in interface GestureModel

Returns:

the saved number of incorrect recognition events.

getNumNotRecognized

public int getNumNotRecognized()

Description copied from interface: GestureModel

Accessor for the number not recognized statistic. This value is entirely dependent on calls to GestureModel.notRecognized()

Specified by:

getNumNotRecognized in interface GestureModel

Returns:

the saved number of unrecognized recognition events.

getNumStates

public int getNumStates()

Returns the number of hidden states in this GestureHMM.

Returns:

the number of hidden states in this GestureHMM.

getProbability

public double getProbability(Vector<Accel3D> instance)

For GestureHMM, probability is computed using the forward algorithm.

Specified by:

getProbability in interface GestureModel

Parameters:

instance - the series of acceleration vectors representing a single gesture instance

Returns:

the probability that the instance matches this model

getTrainingSet

public Vector<Vector<Accel3D>> getTrainingSet()

Description copied from interface: GestureModel

Accessor for the training set of gesture instances associated with this GestureModel

Specified by:

getTrainingSet in interface GestureModel

Returns:

the model's gesture set

getTrainingThreshold

public double getTrainingThreshold()

Returns the training threshold of this GestureHMM.

Returns:

the training threshold of this GestureHMM.

incorrect

public void incorrect()

Description copied from interface: GestureModel

Signifies an incorrect recognition event to the GestureModel. Statistics for the model are updated accordingly.

Specified by:

incorrect in interface GestureModel

notRecognized

public void notRecognized()

Description copied from interface: GestureModel

Signifies an unrecognized recognition event to the GestureModel. Statistics for the model are updated accordingly.

Specified by:

notRecognized in interface GestureModel

removeAccel

public boolean removeAccel(int inum,
                           int anum)

Description copied from interface: GestureModel

Removes an acceleration value from the gesture set.

Specified by:

removeAccel in interface GestureModel

Parameters:

inum - the index of the instance to access

anum - the index of the acceleration to remove

Returns:

a boolean signifying success or failure of the deletion

removeInstance

public boolean removeInstance(int inum)

Description copied from interface: GestureModel

Removes a gesture instance from the gesture set.

Specified by:

removeInstance in interface GestureModel

Parameters:

inum - the index of the instance to remove

Returns:

a boolean signifying success or failure of the deletion

setName

public void setName(String n)

Description copied from interface: GestureModel

Associates a representative name with this GestureModel

Specified by:

setName in interface GestureModel

Parameters:

n - the name to associate

train

public boolean train()

Description copied from interface: GestureModel

Trains the GestureModel, constructing everything necessary for recognition to take place.

Specified by:

train in interface GestureModel

train

public boolean train(GestureHMMParameters gParams)

Trains the GestureHMM with a specified set of training parameters. The training parameters associated with the GestureHMM are set to the new values in gParams.

Parameters:

gParams - the training parameters

Returns:

a boolean indicating whether training took place or not. Training will not occur if the gesture set is empty.

train

public boolean train(GestureHMMParameters gParams,
                     KmeansParameters kParams)

Trains the GestureHMM with a specified set of training parameters as well as a specified set of quantizer parameters. The Quantizer for the GestureHMM is set accordingly and the training parameters associated with the GestureHMM are set to the new values in gParams.

Parameters:

gParams - the training parameters

kParams - the Kmeans parameters

Returns:

a boolean indicating whether training took place or not. Training will not occur if the gesture set is empty.

train

public boolean train(KmeansParameters params)

Trains the GestureHMM with a specified set of quantizer parameters. The Quantizer for the GestureHMM is set accordingly.

Parameters:

params - The K-means parameters.

Returns:

A boolean indicating whether training took place or not. Training will not occur if the gesture set is empty.

setIcon

public void setIcon(byte[] imageBuffer)
             throws IOException

Description copied from interface: GestureModel

Associates a byte array image with the GestureModel

Specified by:

setIcon in interface GestureModel

Parameters:

imageBuffer - the byte array image

Throws:

IOException - thrown if image reading fails

getAverageRecognitionProbability

public double getAverageRecognitionProbability()

Description copied from interface: GestureModel

Returns the average certainty currently associated with this GestureModel. The average certainty is calculated based on all calls made to GestureModel.matchedWithProbability(double).

Specified by:

getAverageRecognitionProbability in interface GestureModel

Returns:

the average recognition probability (certainty)

matchedWithProbability

public void matchedWithProbability(double prob)

Description copied from interface: GestureModel

Signifies a recognition event that occurred with a given probability. This updates the average recognition probability of the GestureModel, as accessed by GestureModel.getAverageRecognitionProbability()

Specified by:

matchedWithProbability in interface GestureModel

Parameters:

prob - the probability (0-1) with which the match occurred

getType

public int getType()

Description copied from interface: GestureModel

Accessor for the GestureModel type index. Should return the value as specified in the GestureModel interface.

Specified by:

getType in interface GestureModel

Returns:

the GestureModel type index.