Home

Awesome

Gesture Recognition Toolkit (GRT)

The Gesture Recognition Toolkit (GRT) is a cross-platform, open-source, C++ machine learning library designed for real-time gesture recognition.

Build Status:

Current version: 0.2.5

Key things to know about the GRT:

Core Resources

Core Algorithms

The GRT supports a wide number of supervised and unsupervised machine learning algorithms for classification, regression, and clustering, including:

In addition to the machine learning algorithms above, the toolkit also includes a large number of algorithms for preprocessing, feature extraction, and post processing.

See the wiki for more details.

GRT Extensions

There are now several extensions and third-party applications that use the GRT as the backend machine learning system, these include:

GRT Architecture

To support flexibility while maintaining consistency, the GRT uses an object-oriented modular architecture. This architecture is built around a set of core modules and a central gesture recognition pipeline.

The input to both the modules and pipeline consists of a N-dimensional floating-point vector, making the toolkit flexible to the type of input signal. The algorithms in each module can be used as standalone classes; alternatively, a pipeline can be used to chain modules together to create a more sophisticated gesture-recognition system. The GRT includes modules for preprocessing, feature extraction, clustering, classification, regression and post processing.

The toolkit's source code is structured as follows:

Getting Started Example

This example demonstrates a few key components of the GRT, such as:

You can find this source code and a large number of other examples and tutorials in the GRT examples folder.

You should run this example with one argument, pointing to the file you want to load, for example:

 ./example my_data.csv

You can find several examples CSV files and other datasets in the main GRT data directory.

//Include the main GRT header
#include <GRT/GRT.h>
using namespace GRT;
using namespace std;

int main (int argc, const char * argv[]) {
  //Parse the training data filename from the command line
  if (argc != 2) {
    cout << "Error: failed to parse data filename from command line. ";
    cout << "You should run this example with one argument pointing to a data file\n";
    return EXIT_FAILURE;
  }
  const string filename = argv[1];

  //Load some training data from a file
  ClassificationData trainingData;

  cout << "Loading dataset..." << endl;
  if (!trainingData.load(filename)) {
    cout << "ERROR: Failed to load training data from file\n";
    return EXIT_FAILURE;
  }

  cout << "Data Loaded" << endl;

  //Print out some stats about the training data
  trainingData.printStats();

  //Partition the training data into a training dataset and a test dataset. 80 means that 80%
  //of the data will be used for the training data and 20% will be returned as the test dataset
  cout << "Splitting data into training/test split..." << endl;
  ClassificationData testData = trainingData.split(80);

  //Create a new Gesture Recognition Pipeline
  GestureRecognitionPipeline pipeline;

  //Add a KNN classifier to the pipeline with a K value of 10
  pipeline << KNN(10);

  //Train the pipeline using the training data
  cout << "Training model..." << endl;
  if (!pipeline.train(trainingData)) {
    cout << "ERROR: Failed to train the pipeline!\n";
    return EXIT_FAILURE;
  }

  //Save the pipeline to a file
  if (!pipeline.save("HelloWorldPipeline.grt")) {
    cout << "ERROR: Failed to save the pipeline!\n";
    return EXIT_FAILURE;
  }

  //Load the pipeline from a file
  if (!pipeline.load("HelloWorldPipeline.grt")) {
    cout << "ERROR: Failed to load the pipeline!\n";
    return EXIT_FAILURE;
  }

  //Test the pipeline using the test data
  cout << "Testing model..." << endl;
  if (!pipeline.test(testData)) {
    cout << "ERROR: Failed to test the pipeline!\n";
    return EXIT_FAILURE;
  }

  //Print some stats about the testing
  cout << "Pipeline Test Accuracy: " << pipeline.getTestAccuracy() << endl;

  //Manually project the test dataset through the pipeline
  Float testAccuracy = 0.0;
  for (UINT i=0; i<testData.getNumSamples(); i++) {
    pipeline.predict(testData[i].getSample());

    if (testData[i].getClassLabel() == pipeline.getPredictedClassLabel()) {
      testAccuracy++;
    }
  }
  cout << "Manual test accuracy: " << testAccuracy / testData.getNumSamples() * 100.0 << endl;
   
  //Get the vector of class labels from the pipeline
  Vector< UINT > classLabels = pipeline.getClassLabels();

  //Print out the precision
  cout << "Precision: ";
  for (UINT k=0; k<pipeline.getNumClassesInModel(); k++) {
    cout << "\t" << pipeline.getTestPrecision(classLabels[k]);
  }cout << endl;

  //Print out the recall
  cout << "Recall: ";
  for (UINT k=0; k<pipeline.getNumClassesInModel(); k++) {
    cout << "\t" << pipeline.getTestRecall(classLabels[k]);
  }cout << endl;

  //Print out the f-measure
  cout << "FMeasure: ";
  for (UINT k=0; k<pipeline.getNumClassesInModel(); k++) {
    cout << "\t" << pipeline.getTestFMeasure(classLabels[k]);
  }cout << endl;

  //Print out the confusion matrix
  MatrixFloat confusionMatrix = pipeline.getTestConfusionMatrix();
  cout << "ConfusionMatrix: \n";
  for (UINT i=0; i<confusionMatrix.getNumRows(); i++) {
    for (UINT j=0; j<confusionMatrix.getNumCols(); j++) {
      cout << confusionMatrix[i][j] << "\t";
    }cout << endl;
  }

  return EXIT_SUCCESS;
}

Tutorials and Examples

You can find a large number of tutorials and examples in the examples folder. You can also find a wide range of examples and references on the main GRT wiki:

http://www.nickgillian.com/wiki/pmwiki.php?n=GRT.GestureRecognitionToolkit

If you build the GRT using CMake, an examples folder will automatically be generated in the build directory after you successfully build the main GRT library. Example applications can then be directly run from this example directory. To run any of the examples, open terminal in the grt/build/examples directory and run:

./ExampleName

where ExampleName is the name of the example application you want to run.

Forum

Note, at the moment the forum server is currently broken, we are working to resolve this. In the meantime, use GitHub issues and pull requests.

You can find the link for the old forum at: http://www.nickgillian.com/forum/

Bugs

Please submit bugs to the github bug tracker.

Contributions

All contributions are welcome, there are several ways in which users can contribute to the toolkit:

Please submit pull requests for any contribution.

GRT Floating Point Precision

The GRT defaults to double precision floating point values. The precision of the toolkit is defined by the following Float typedef:

typedef double Float; ///<This typedef is used to set floating-point precision throughout the GRT

This can easily be changed to single precision accuracy if needed by modifying the main GRT Float typedef value, defined in GRT/Util/GRTTypedefs.h header.

VectorFloat and MatrixFloat Data Structures

The GRT uses two main data structures throughout the toolkit: Vector and Matrix. These are templates and can, therefore, generalize to any C++ class. The main things to know about these data types are:

//Create an integer vector with a size of 3 elements
Vector< int > vec1(3);

//Create a string vector with a size of 2 elements
Vector< string > vec2(2);

//Create a Foo vector with a size of 5 elements
Vector< Foo > vec3(5);

//Create an integer matrix with a size of 3x2
Matrix< int > mat1(3,2);

//Create a string matrix with a size of 2x2
Matrix< string > mat2(2,2);

//Create a Foo matrix with a size of 5x3
Matrix< Foo > mat3(5,3);

//Create a new vector with 10 elements
VectorFloat vector( 10 );
for(UINT i=0; i<vector.getSize(); i++){ 
    vector[i] = i*1.0; 
}

//Create a [5x2] floating point matrix
MatrixFloat matrix(5,2);

//Loop over the data and set the values to a basic incrementing value
UINT counter = 0;
for(UINT i=0; i<matrix.getNumRows(); i++){
    for(UINT j=0; j<matrix.getNumCols(); j++){
        matrix[i][j] = counter++;
    }
}

Building the GRT

You can find a CMakeLists file in the build folder that you can use to auto generate a makefile for your machine.

Read the readme file in the build folder to see how to build the GRT as a static library for Linux, OS X, or Windows.

Installing and using the GRT in your C++ projects

See the build directory for details on how to build, install, and use the GRT in your C++ projects.

License

The Gesture Recognition Toolkit is available under an MIT license.

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.