Gesture Models for Multitouch

/*

 * Gestures.h

 *

 *  Created on: Jun 8, 2009

 *      Author: damaraju

 */

#ifndef GESTURES_H_

#define GESTURES_H_

#include <map>

#include <string>

#include <fstream>

#include <touch/Touch.h>

#include <gesture/models/VectorGestureClassification.h>

#include <boost/range/adaptor/transformed.hpp>

#include <boost/range/algorithm/copy.hpp>

using namespace std;

//For debug and test

string base = "data/reOrdered/";

void printTransform(vector<vector<double> > transformed)

{

        cout << "Transformed Size: " << transformed.size() << endl;

        for(size_t i = 0; i < transformed.size(); ++i)

        {

                for (size_t j = 0; j < transformed.at(i).size(); ++j)

                        cout << i << ": " << transformed.at(i).at(j) << "\t";

                cout << endl;

        }

}

vector<GestureSample> readFile(string fileName)

{

        vector<GestureSample> readSamples;

        ifstream readFile;

        cout << "Reading: " << fileName << " " << endl;

        readFile.open(fileName.c_str());

        string line;

        if (readFile.is_open())

        {

                while(!readFile.eof())

                {

                        getline(readFile,line);

                        GestureSample sample = GestureSample(line);

                        if(sample.size() > 0)

                                readSamples.push_back(sample);

                }

                cout << "Num Samples: " << readSamples.size() << endl;

                readFile.close();

        }

        else

        {

                cout << "Unable to read file: " << fileName << endl;

        }

        return readSamples;

}

vector<GestureSample> readGestureSet(string gid, string uid)

{

        string fileName = "gid_" + gid + "_uid_" + uid + ".seqs";

        return readFile(base + fileName);

}

vector<vector<vector<double> > > transformSamples(vector<GestureSample> samples)

{

        vector<vector<vector<double> > > transformedSet;

        //cout << "Transforming " << samples.size() << " samples" << endl;

        if (samples.size() > 0)

                for (size_t i = 0; i < samples.size(); ++i)

                        transformedSet.push_back(samples.at(i).transform());

        else

                cout << "No Samples" << endl;

        return transformedSet;

}

class RecognitionHelper

{

public:

        VectorGestureClassification classifier;

        map<int, string>                         gestureNameMap;

        RecognitionHelper()

        {}

        void trainWithSamples(vector<GestureSample> trainingSet, string gestureName)

        {

                vector<vector<vector<double> > > trnsfTrain = transformSamples(trainingSet);

                //Magical filter takes care of all preprocessing, and live processing of the incoming sample(s).

                multitouch_filter filter = multitouch_filter(trnsfTrain);

                //Ensure the training set is appropriately scaled for training the model.

                vector<vector<vector<double> > > filteredTraining = trnsfTrain;

                for(size_t i = 0; i < trnsfTrain.size(); i++)

                {

                        filter.reset_params_for(trnsfTrain[i]);

                        boost::copy(trnsfTrain[i] | boost::adaptors::transformed(filter),filteredTraining[i].begin());

                }

                cout << "Training With: " << filteredTraining.size() << " samples" << endl;

                classifier.addGestureWithExamplesAndFilter(filteredTraining, 5, filter);

                gestureNameMap.insert(pair<int, string>(classifier.numGestures() - 1, gestureName));

                cout << "\nAdded new gesture(" << classifier.numGestures() - 1 << "): " << gestureName << "\n"<<endl;

        }

        string classify(GestureSample sample)

        {

                vector<vector<double> > transformed = sample.transform();

                //printTransform(transformed);

                int classIndex = classifier.classify(transformed);

                vector<long double> probs = classifier.probabilities();

                bool allZero = true;

                for(unsigned int i = 0 ; i < probs.size(); i++)

                {

                        if(probs[i] > 0)

                                allZero = false;

                        cout << probs[i] << ", ";

                }

                cout << endl;

                if(!allZero && classIndex >= 0 && ((unsigned int)classIndex) < gestureNameMap.size())

                        return gestureNameMap[classIndex];

                else //classIndex == -1 || allZero probabilities when sample doesn't match any filter-model pair

                        return "None";

        }

        const std::vector<long double> &probabilities() const

        {

            return classifier.probabilities();

        }

};

#endif /* GESTURES_H_ */