void FeatureEval::pca(){
boost::shared_ptr<PointCloud> cloud = core_->cl_->active_;
if(cloud == nullptr)
return;
std::vector<int> sub_idxs;
pcl::PointCloud<pcl::PointXYZI>::Ptr smaller_cloud = octreeDownsample(cloud.get(), subsample_res_, sub_idxs);
t_.start(); qDebug() << "Timer started (PCA)";
boost::shared_ptr<std::vector<Eigen::Vector3f> > pca = getPCA(smaller_cloud.get(), search_radius_, max_nn_);
qDebug() << "PCA in " << (time_ = t_.elapsed()) << " ms";
computeCorrelation(reinterpret_cast<float *>(pca->data()), 3, pca->size(), sub_idxs, sizeof(Eigen::Vector3f)/sizeof(float));
if(!visualise_on_) return;
// // Draw
// boost::shared_ptr<std::vector<Eigen::Vector3f> > grid = boost::make_shared<std::vector<Eigen::Vector3f> >(grid_to_cloud->size(), Eigen::Vector3f(0.0f, 0.0f, 0.0f));
// for(int i = 0; i < grid_to_cloud->size(); i++) {
// int idx = (*grid_to_cloud)[i];
// if(idx != -1)
// (*grid)[i] = (*pca)[idx];
// }
// drawVector3f(grid, cloud);
}
void FeatureEval::eigen_plane_consine_similarity(){
boost::shared_ptr<PointCloud> cloud = core_->cl_->active_;
if(cloud == nullptr)
return;
std::vector<int> sub_idxs;
pcl::PointCloud<pcl::PointXYZI>::Ptr smaller_cloud = octreeDownsample(cloud.get(), subsample_res_, sub_idxs);
t_.start(); qDebug() << "Timer started (eigen_plane_consine_similarity)";
boost::shared_ptr<std::vector<Eigen::Vector3f> > pca = getPCA(smaller_cloud.get(), 0.05f, 20);
boost::shared_ptr<std::vector<float>> plane_likelyhood =
boost::make_shared<std::vector<float>>(pca->size(), 0.0f);
Eigen::Vector3f ideal_plane(1.0f, 0.0f, 0.0f);
ideal_plane.normalize();
for(uint i = 0; i < pca->size(); i++) {
Eigen::Vector3f & val = (*pca)[i];
// Not enough neighbours
if(val[1] < val[2]) {
(*plane_likelyhood)[i] = 0;
continue;
}
float similarity = cosine(val, ideal_plane);
(*plane_likelyhood)[i] = similarity;
}
qDebug() << "eigen_plane_consine_similarity in " << (time_ = t_.elapsed()) << " ms";
// Correlate
computeCorrelation(plane_likelyhood->data(), 1, plane_likelyhood->size(), sub_idxs);
if(!visualise_on_) return;
// Draw
int h = cloud->scan_width();
int w = cloud->scan_height();
/*
boost::shared_ptr<std::vector<Eigen::Vector3f> > grid = boost::make_shared<std::vector<Eigen::Vector3f> >(grid_to_cloud->size(), Eigen::Vector3f(0.0f, 0.0f, 0.0f));
for(int i = 0; i < grid_to_cloud->size(); i++) {
int idx = (*grid_to_cloud)[i];
if(idx != -1)
(*grid)[i] = (*pca)[idx];
}
drawVector3f(grid, cloud);
*/
boost::shared_ptr<const std::vector<float>> img = cloudToGrid(cloud->cloudToGridMap(), w*h, plane_likelyhood);
drawFloats(img, cloud);
}
void TLinearCombBoxCox::calcSimDataPCA(double* simData){
//first transform stats via boxcox
int stat=0;
for(int i=0; i<numStats; ++i){
if(statIsUsed[i]>=0){
simData[i]=getBoxCox(simData[i], stat);
++stat;
}
}
for(int i=0; i<numLinearComb; ++i){
simPCA[i]=getPCA(i, simData);
}
}
//------------------------------------------------------------------------------
void TLinearCombBoxCox::calcObsLineraComb(vector<double> obsData){
//order has to be the same as the names passed to the constructor!!!
obs=new double[numStats];
for(int i=0; i<numStats; ++i){
if(statIsUsed[i]>=0){
obs[statIsUsed[i]]=getBoxCox(obsData[i], i);;
}
}
//calculate pca for obsData
obsPCA=new double[numLinearComb];
for(int i=0; i<numLinearComb; ++i){
obsPCA[i]=getPCA(i, obs);
}
}
// PCA for each point, ratio of eigen values
void FeatureEval::pca_eigen_value_ratio(){
boost::shared_ptr<PointCloud> cloud = core_->cl_->active_;
if(cloud == nullptr)
return;
// Downsample
std::vector<int> sub_idxs;
pcl::PointCloud<pcl::PointXYZI>::Ptr smaller_cloud = octreeDownsample(cloud.get(), subsample_res_, sub_idxs);
// Compute PCA
t_.start(); qDebug() << "Timer started (PCA and speculaion)";
boost::shared_ptr<std::vector<Eigen::Vector3f> > pca = getPCA(smaller_cloud.get(), search_radius_, max_nn_);
// Compute ratio
boost::shared_ptr<std::vector<float>> likely_veg =
boost::make_shared<std::vector<float>>(pca->size(), 0.0f);
for(uint i = 0; i < pca->size(); i++) {
Eigen::Vector3f eig = (*pca)[i];
// Not enough neighbours
if(eig[1] < eig[2]) {
(*likely_veg)[i] = 0;
continue;
}
/*
float eig_sum = eig.sum();
eig /= eig_sum;
float fudge_factor = 5.0f;
bool is_planar = eig[1] < 0.05 * fudge_factor || eig[2] < 0.01 * fudge_factor;
if(!is_planar) {
(*likely_veg)[i] = 0;
} else {
(*likely_veg)[i] = 1;
}
*/
float curvature;
if(eig[0] > eig[2])
curvature = eig[1] / (eig[0]);
else
curvature = 1;
(*likely_veg)[i] = curvature;
}
qDebug() << "PCA in " << (time_ = t_.elapsed()) << " ms";
computeCorrelation(likely_veg->data(), 1, likely_veg->size(), sub_idxs);
if(!visualise_on_) return;
// Draw
int h = cloud->scan_width();
int w = cloud->scan_height();
boost::shared_ptr<std::vector<float>> likely_veg2 =
boost::make_shared<std::vector<float>>(cloud->size(), 0.0f);
for(uint i = 0; i < cloud->size(); i++) {
uint subidx = sub_idxs[i];
(*likely_veg2)[i] = (*likely_veg)[subidx];
}
boost::shared_ptr<const std::vector<float>> img = cloudToGrid(cloud->cloudToGridMap(), w*h, likely_veg2);
drawFloats(img, cloud);
}
