void Camera::init(const mo3d::NVM_Camera* cam, int width, int height, const int maxLevel){
projection_.resize(maxLevel+1);
kMat_.resize(maxLevel+1);
// calculate projection matrix for level 0
kMat_[0] << cam->f , 0 , width/2.0 , 0 , cam->f , height/2.0 , 0, 0, 1;
Eigen::Quaterniond q;
q.w() = cam->rq[0];
q.x() = cam->rq[1];
q.y() = cam->rq[2];
q.z() = cam->rq[3];
projection_[0].col(3) = -q.matrix().cast<float>() * cam->c.cast<float>() ;
projection_[0].topLeftCorner(3,3) = q.matrix().cast<float>();
projection_[0] = kMat_[0] * projection_[0];
// now propagate to lower levels:
for (int l = 1; l < maxLevel+1; l++){
projection_[l] = projection_[l-1];
projection_[l].row(0) /= 2.0;
projection_[l].row(1) /= 2.0;
kMat_[l] = kMat_[l-1];
kMat_[l].row(0) /= 2.0;
kMat_[l].row(1) /= 2.0;
}
// optical center
center_.head(3) = cam->c.cast<float>();
center_(3) = 1;
// axis of the camera
oAxis_ = projection_[0].row(2);
oAxis_ /= projection_[0].row(2).head(3).norm();
zAxis_ = oAxis_.head(3);
xAxis_ = projection_[0].row(0).head(3);
yAxis_ = zAxis_.cross(xAxis_).normalized();
xAxis_ = yAxis_.cross(zAxis_).normalized();
// get the average pixelscale
ipscale_ = (projection_[0].row(0).head(3).norm() + projection_[0].row(1).head(3).norm())/2.0;
}
void
MFCC::init()
{
unsigned int i,j;
unsigned int chan;
if (inSize_ != DEFAULT_WIN_SIZE)
{
cerr << "MFCC::init warning inSize_ != 512\n" << endl ;
}
// Initialize frequency boundaries for the filters
freqs_.create(42);
lowestFrequency_ = 133.3333f;
linearFilters_ = 13;
linearSpacing_ = 66.66666f;
logFilters_ = 27;
logSpacing_ = 1.0711703f;
totalFilters_ = linearFilters_ + logFilters_;
lower_.create(totalFilters_);
center_.create(totalFilters_);
upper_.create(totalFilters_);
triangle_heights_.create(totalFilters_);
// Linear filter boundaries
for (i=0; i< linearFilters_; i++)
freqs_(i) = lowestFrequency_ + i * linearSpacing_;
// Logarithmic filter boundaries
float first_log = freqs_(linearFilters_-1);
for (i=1; i<=logFilters_+2; i++)
{
freqs_(linearFilters_-1+i) = first_log * pow(logSpacing_, (float)i);
}
// Triangles information
for (i=0; i<totalFilters_; i++)
lower_(i) = freqs_(i);
for (i=1; i<= totalFilters_; i++)
center_(i-1) = freqs_(i);
for (i=2; i<= totalFilters_+1; i++)
upper_(i-2) = freqs_(i);
for (i=0; i<totalFilters_; i++)
triangle_heights_(i) = 2.0 / (upper_(i) - lower_(i));
fftSize_ = 512;
samplingRate_ = 22050;
fftFreqs_.create(fftSize_);
cepstralCoefs_ = 13;
for (i=0; i< fftSize_; i++)
fftFreqs_(i) = (float)i / (float)fftSize_ * (float)samplingRate_;
mfccFilterWeights_.create(totalFilters_, fftSize_);
mfccDCT_.create(cepstralCoefs_, totalFilters_);
// Initialize mfccFilterWeights
for (chan = 0; chan < totalFilters_; chan++)
for (i=0; i< fftSize_; i++)
{
if ((fftFreqs_(i) > lower_(chan))&& (fftFreqs_(i) <= center_(chan)))
{
mfccFilterWeights_(chan, i) = triangle_heights_(chan) *
((fftFreqs_(i) - lower_(chan))/(center_(chan) - lower_(chan)));
}
if ((fftFreqs_(i) > center_(chan)) && (fftFreqs_(i) <= upper_(chan)))
{
mfccFilterWeights_(chan, i) = triangle_heights_(chan) *
((upper_(chan) - fftFreqs_(i))/(upper_(chan) - center_(chan)));
}
}
// Initialize MFCC_DCT
float scale_fac = 1.0/ sqrt((double)(totalFilters_/2));
for (j = 0; j<cepstralCoefs_; j++)
for (i=0; i< totalFilters_; i++)
{
mfccDCT_(j, i) = scale_fac * cos(j * (2*i +1) * PI/2/totalFilters_);
if (i == 0)
mfccDCT_(j,i) *= sqrt(2.0)/2.0;
}
// Prepare feature names
featSize_ = outSize_;
featNames_.push_back("MFCC00");
featNames_.push_back("MFCC01");
featNames_.push_back("MFCC02");
featNames_.push_back("MFCC03");
featNames_.push_back("MFCC04");
featNames_.push_back("MFCC05");
featNames_.push_back("MFCC06");
featNames_.push_back("MFCC07");
featNames_.push_back("MFCC08");
featNames_.push_back("MFCC09");
featNames_.push_back("MFCC10");
featNames_.push_back("MFCC11");
featNames_.push_back("MFCC12");
hamming_ = new Hamming(inSize_, zeroSize_);
magfft_ = new MagFFT(inSize_);
windowed.create(inSize_);
magnitude.create(inSize_/2);
fmagnitude.create(inSize_);
earMagnitude_.create(totalFilters_);
}
uint SelectSubvolume(vec_t point) {
std::bitset<D> bitindex;
for(int i=0;i<D;++i)
bitindex[i] = point(i) > center_(i) ? 1 : 0;
return bitindex.to_ulong();
}
double center(int i) const { return center_(i); }
