UINT GesturesAnalyzer::PredictGesture(const vector<vector<cv::Point3d>>& pSampleVec)
{
GRT::MatrixDouble sample = GRT::MatrixDouble( );
vector<double> vecTemp;
UINT predictedClassLabel = 0;
for(int iterator = 0; iterator < pSampleVec.size(); iterator++)
{
int x0 = pSampleVec[iterator][nite::JOINT_TORSO].x;
int y0 = pSampleVec[iterator][nite::JOINT_TORSO].y;
int z0 = pSampleVec[iterator][nite::JOINT_TORSO].z;
vecTemp.push_back(pSampleVec[iterator][nite::JOINT_RIGHT_HAND].x - x0);
vecTemp.push_back(pSampleVec[iterator][nite::JOINT_RIGHT_HAND].y - y0);
vecTemp.push_back(pSampleVec[iterator][nite::JOINT_RIGHT_HAND].z - z0);
vecTemp.push_back(pSampleVec[iterator][nite::JOINT_LEFT_HAND].x - x0);
vecTemp.push_back(pSampleVec[iterator][nite::JOINT_LEFT_HAND].y - y0);
vecTemp.push_back(pSampleVec[iterator][nite::JOINT_LEFT_HAND].z - z0);
sample.push_back(vecTemp);
vecTemp.clear();
}
if( !m_dtw.predict(sample))
{
printf("Failed prediction !\n");
}
else
{
predictedClassLabel = m_dtw.getPredictedClassLabel();
}
return predictedClassLabel;
}
void HiddenMarkovModels::setUpTrainingDataset() {
/* Setup Quantizer */
if (quantizer != NULL) {
delete(quantizer);
}
quantizer = new GRT::KMeansQuantizer( trainingDataset.getNumDimensions(), NUM_SYMBOLS );
//Train the quantizer using the training data
if(!quantizer->train( trainingDataset ) ){
throw TestModelException("HiddenMarkovModels: ERROR: Failed to train quantizer!");
}
//Quantize the training data
GRT::LabelledTimeSeriesClassificationData quantizedTrainingData( 1 );
for( GRT::UINT i=0; i<trainingDataset.getNumSamples(); i++){
GRT::UINT classLabel = trainingDataset[i].getClassLabel();
GRT::MatrixDouble quantizedSample;
for(GRT::UINT j=0; j<trainingDataset[i].getLength(); j++) {
quantizer->quantize( trainingDataset[i].getData().getRowVector(j) );
quantizedSample.push_back( quantizer->getFeatureVector() );
}
if( !quantizedTrainingData.addSample(classLabel, quantizedSample) ){
throw TestModelException("HiddenMarkovModels: ERROR: Failed to quantize training data!");
}
}
this->trainingDataset = quantizedTrainingData;
}
void ASCIISerialStream::readSerial() {
// TODO(benzh) This readSerial is running in a different thread
// and performing a busy polling (100% CPU usage). Should be
// optimized.
int sleep_time = 10;
ofLog() << "Serial port will be read every " << sleep_time << " ms";
while (has_started_) {
string s;
do {
while (serial_->available() < 1) {
std::this_thread::sleep_for(std::chrono::milliseconds(sleep_time));
}
s += serial_->readByte();
} while(s[s.length() - 1] != '\n');
//ofLog() << "read: '" << s << "'" << endl;
if (data_ready_callback_ != nullptr) {
istringstream iss(s);
vector<double> data;
double d;
while (iss >> d) data.push_back(d);
if (data.size() > 0) {
data = normalize(data);
GRT::MatrixDouble matrix;
matrix.push_back(data);
data_ready_callback_(matrix);
}
}
}
HRESULT GesturesAnalyzer::DTWSaveSampleAndTrain(UINT pClasse, const vector<vector<cv::Point3d>>& pSampleVec)
{
printf("Start saving DTW sample\n");
GRT::MatrixDouble sample = GRT::MatrixDouble( );
vector<double> vecTemp;
for(int iterator = 0; iterator < pSampleVec.size(); iterator++)
{
int x0 = pSampleVec[iterator][nite::JOINT_TORSO].x;
int y0 = pSampleVec[iterator][nite::JOINT_TORSO].y;
int z0 = pSampleVec[iterator][nite::JOINT_TORSO].z;
vecTemp.push_back(pSampleVec[iterator][nite::JOINT_RIGHT_HAND].x - x0);
vecTemp.push_back(pSampleVec[iterator][nite::JOINT_RIGHT_HAND].y - y0);
vecTemp.push_back(pSampleVec[iterator][nite::JOINT_RIGHT_HAND].z - z0);
vecTemp.push_back(pSampleVec[iterator][nite::JOINT_LEFT_HAND].x - x0);
vecTemp.push_back(pSampleVec[iterator][nite::JOINT_LEFT_HAND].y - y0);
vecTemp.push_back(pSampleVec[iterator][nite::JOINT_LEFT_HAND].z - z0);
sample.push_back(vecTemp);
vecTemp.clear();
}
printf("Adding sample to DTW data\n");
if( !m_DTWTrainingData.addSample(pClasse, sample))
{
printf("Failed to add DTW sample\n");
return E_FAIL;
}
if( !m_DTWTrainingData.saveDatasetToFile("TrainingData\\DTWTrainingData.txt"))
{
printf("Failed to save DTW data\n");
return E_FAIL;
}
printf("Training DTW classifier\n");
if( !m_dtw.train( m_DTWTrainingData ) ){
cout << "Failed to train DTW classifier!\n";
return E_FAIL;
}
else
{
printf("Succesfully trained DTW Classifier\n");
}
return S_OK;
}
void HiddenMarkovModels::setUpTestingDataset() {
//Quantize the test data
GRT::LabelledTimeSeriesClassificationData quantizedTestData( 1 );
for(GRT::UINT i=0; i < testDataset.getNumSamples(); i++){
GRT::UINT classLabel = testDataset[i].getClassLabel();
GRT::MatrixDouble quantizedSample;
for(GRT::UINT j=0; j<testDataset[i].getLength(); j++){
quantizer->quantize( testDataset[i].getData().getRowVector(j) );
quantizedSample.push_back( quantizer->getFeatureVector() );
}
if( !quantizedTestData.addSample(classLabel, quantizedSample) ){
throw TestModelException("HiddenMarkovModels: ERROR: Failed to quantize training data!");
}
}
this->testDataset = quantizedTestData;
}
void ASCIISerialStream::parseSerial(vector<unsigned char> &buffer) {
auto newline = find(buffer.begin(), buffer.end(), '\n');
if (newline != buffer.end()) {
string s(buffer.begin(), newline);
buffer.erase(buffer.begin(), ++newline);
if (data_ready_callback_ != nullptr) {
istringstream iss(s);
vector<double> data;
double d;
while (iss >> d) data.push_back(d);
if (data.size() > 0) {
data = normalize(data);
GRT::MatrixDouble matrix;
matrix.push_back(data);
data_ready_callback_(matrix);
}
}