3dpcp/.svn/pristine/b1/b139234db708baffd361e6e7f0fa8d88aaf822e5.svn-base
2012-11-13 16:49:04 +01:00

134 lines
3.9 KiB
Text

#ifndef NORMALS_H
#define NORMALS_H
#include <vector>
#include <slam6d/scan.h>
#if (CV_MAJOR_VERSION == 2) && (CV_MINOR_VERSION < 2)
#include <opencv/cv.h>
#include <opencv/highgui.h>
#else
#include <opencv2/opencv.hpp>
#endif
void calculateNormalsAKNN(std::vector<Point> &normals,vector<Point> &points, int k,
const double _rPos[3] );
void calculateNormalsAdaptiveAKNN(vector<Point> &normals,vector<Point> &points,
int kmin, int kmax, const double _rPos[3]);
void calculateNormalsPANORAMA(vector<Point> &normals,
vector<Point> &points,
vector< vector< vector< cv::Vec3f > > > extendedMap,
const double _rPos[3]);
// TODO should be exported to separate library
/*
* retrieve a cv::Mat with x,y,z,r from a scan object
* functionality borrowed from scan_cv::convertScanToMat but this function
* does not allow a scanserver to be used, prints to stdout and can only
* handle a single scan
*/
static inline cv::Mat scan2mat(Scan *source)
{
DataXYZ xyz = source->get("xyz");
DataReflectance xyz_reflectance = source->get("reflectance");
unsigned int nPoints = xyz.size();
cv::Mat scan(nPoints,1,CV_32FC(4));
scan = cv::Scalar::all(0);
cv::MatIterator_<cv::Vec4f> it;
it = scan.begin<cv::Vec4f>();
for(unsigned int i = 0; i < nPoints; i++){
float x, y, z, reflectance;
x = xyz[i][0];
y = xyz[i][1];
z = xyz[i][2];
if(xyz_reflectance.size() != 0)
{
reflectance = xyz_reflectance[i];
//normalize the reflectance
reflectance += 32;
reflectance /= 64;
reflectance -= 0.2;
reflectance /= 0.3;
if (reflectance < 0) reflectance = 0;
if (reflectance > 1) reflectance = 1;
}
(*it)[0] = x;
(*it)[1] = y;
(*it)[2] = z;
if(xyz_reflectance.size() != 0)
(*it)[3] = reflectance;
else
(*it)[3] = 0;
++it;
}
return scan;
}
// TODO should be exported to separate library
/*
* convert a matrix of float values (range image) to a matrix of unsigned
* eight bit characters using different techniques
*/
static inline cv::Mat float2uchar(cv::Mat &source, bool logarithm, float cutoff)
{
cv::Mat result(source.size(), CV_8U, cv::Scalar::all(0));
float max = 0;
// find maximum value
if (cutoff == 0.0) {
// without cutoff, just iterate through all values to find the largest
for (cv::MatIterator_<float> it = source.begin<float>();
it != source.end<float>(); ++it) {
float val = *it;
if (val > max) {
max = val;
}
}
} else {
// when a cutoff is specified, sort all the points by value and then
// specify the max so that <cutoff> values are larger than it
vector<float> sorted(source.cols*source.rows);
int i = 0;
for (cv::MatIterator_<float> it = source.begin<float>();
it != source.end<float>(); ++it, ++i) {
sorted[i] = *it;
}
std::sort(sorted.begin(), sorted.end());
max = sorted[(int)(source.cols*source.rows*(1.0-cutoff))];
cout << "A cutoff of " << cutoff << " resulted in a max value of " << max << endl;
}
cv::MatIterator_<float> src = source.begin<float>();
cv::MatIterator_<uchar> dst = result.begin<uchar>();
cv::MatIterator_<float> end = source.end<float>();
if (logarithm) {
// stretch values from 0 to max logarithmically over 0 to 255
// using the logarithm allows to represent smaller values with more
// precision and larger values with less
max = log(max+1);
for (; src != end; ++src, ++dst) {
float val = (log(*src+1)*255.0)/max;
if (val > 255)
*dst = 255;
else
*dst = (uchar)val;
}
} else {
// stretch values from 0 to max linearly over 0 to 255
for (; src != end; ++src, ++dst) {
float val = (*src*255.0)/max;
if (val > 255)
*dst = 255;
else
*dst = (uchar)val;
}
}
return result;
}
#endif // NORMALS_H