3dpcp/src/slam6d/icp6D.cc

/*
 * icp6D implementation
 *
 * Copyright (C) Andreas Nuechter, Kai Lingemann
 *
 * Released under the GPL version 3.
 *
 */


/**
 * @file 
 * @brief Implementation of 3D scan matching with ICP
 * @author Kai Lingemann. Institute of Computer Science, University of Osnabrueck, Germany.
 * @author Andreas Nuechter. Institute of Computer Science, University of Osnabrueck, Germany.
 */

#include "slam6d/icp6D.h"

#include "slam6d/metaScan.h"
#include "slam6d/globals.icc"

#include <iomanip>
using std::cerr;

#include <string.h>

#ifdef _MSC_VER
#if !defined _OPENMP && defined OPENMP 
#define _OPENMP
#endif
#endif

#ifdef _OPENMP
#include <omp.h>
#endif
/**
 * Constructor 
 *
 * @param my_icp6Dminimizer Pointer to the ICP-minimizer
 * @param max_dist_match Maximum distance to which point pairs are collected
 * @param max_num_iterations Maximum number of iterations
 * @param quiet Whether to print to the standard output
 * @param meta Match against a meta scan?
 * @param rnd Randomized point selection
 * @param eP Extrapolate odometry?
 * @param anim Animate which frames?
 * @param epsilonICP Termination criterion
 * @param nns_method Selects NNS method to be used  
 */
icp6D::icp6D(icp6Dminimizer *my_icp6Dminimizer, double max_dist_match,
		   int max_num_iterations, bool quiet, bool meta, int rnd, bool eP,
		   int anim, double epsilonICP, int nns_method, bool cuda_enabled,
       bool cad_matching)
{
  this->my_icp6Dminimizer = my_icp6Dminimizer;
  this->anim              = anim;
  this->cuda_enabled      = cuda_enabled;
  this->nns_method        = nns_method;
  
  if (!quiet) {
    cout << "Maximal distance match      : " << max_dist_match << endl
	 << "Maximal number of iterations: " << max_num_iterations << endl << endl;
  }
  
  // checks
  if (max_dist_match < 0.0) {
    cerr << "ERROR [ICP6D]: first parameter (max_dist_match) has to be >= 0," << endl;
    exit(1);
  }
  if (max_num_iterations < 0) {
    cerr << "ERROR [ICP6D]: second parameter (max_num_iterations) has to be >= 0." << endl;
    exit(1);
  }
  
  this->max_dist_match2    = sqr(max_dist_match);
  this->max_num_iterations = max_num_iterations;
  this->quiet              = quiet;
  this->meta               = meta;
  this->rnd                = rnd;
  this->eP                 = eP;
  this->epsilonICP         = epsilonICP;
  
  // Set initial seed (for "real" random numbers)
  //  srand( (unsigned)time( NULL ) );
  this->cad_matching = cad_matching;
}

/**
 * Matches a 3D Scan against a 3D Scan
 * @param PreviousScan The scan or metascan forming the model
 * @param CurrentScan The current scan thas is to be matched
 * @return The number of iterations done in this matching run
 */
int icp6D::match(Scan* PreviousScan, Scan* CurrentScan,
                 PairingMode pairing_mode)
{
  // If ICP shall not be applied, then just write
  // the identity matrix
  if (max_num_iterations == 0) {
    double id[16];
    M4identity(id);
    CurrentScan->transform(id, Scan::ICP, 0);  // write end pose
    return 0;
  }

  // icp main loop
  double ret = 0.0, prev_ret = 0.0, prev_prev_ret = 0.0;
  int iter = 0;
  double alignxf[16];
  for (iter = 0; iter < max_num_iterations; iter++) {

    prev_prev_ret = prev_ret;
    prev_ret = ret;

#ifdef _OPENMP
    // Implementation according to the paper 
    // "The Parallel Iterative Closest Point Algorithm"
    // by Langis / Greenspan / Godin, IEEE 3DIM 2001
    //
    // The same information are given in (ecrm2007.pdf)
    // Andreas Nüchter. Parallelization of Scan Matching
    // for Robotic 3D Mapping. In Proceedings of the 3rd
    // European Conference on Mobile Robots (ECMR '07),
    // Freiburg, Germany, September 2007
    omp_set_num_threads(OPENMP_NUM_THREADS);

    int max = (int)CurrentScan->size<DataXYZ>("xyz reduced");
    int step = max / OPENMP_NUM_THREADS;

    vector<PtPair> pairs[OPENMP_NUM_THREADS];
    double sum[OPENMP_NUM_THREADS];
    double centroid_m[OPENMP_NUM_THREADS][3];
    double centroid_d[OPENMP_NUM_THREADS][3];
    double Si[OPENMP_NUM_THREADS][9];
    unsigned int n[OPENMP_NUM_THREADS];

    for (int i = 0; i < OPENMP_NUM_THREADS; i++) {
      sum[i] = centroid_m[i][0] = centroid_m[i][1] = centroid_m[i][2] = 0.0;
      centroid_d[i][0] = centroid_d[i][1] = centroid_d[i][2] = 0.0;
      Si[i][0] = Si[i][1] = Si[i][2] = Si[i][3] = Si[i][4] = Si[i][5] = Si[i][6] = Si[i][7] = Si[i][8] = 0.0;
      n[i] = 0;
    }

#pragma omp parallel 
    {
      int thread_num = omp_get_thread_num();

      Scan::getPtPairsParallel(pairs, PreviousScan, CurrentScan,
          thread_num, step,
          rnd, max_dist_match2,
          sum, centroid_m, centroid_d, pairing_mode);

      n[thread_num] = (unsigned int)pairs[thread_num].size();

      if ((my_icp6Dminimizer->getAlgorithmID() == 1) ||
          (my_icp6Dminimizer->getAlgorithmID() == 2)) {
        for (unsigned int i = 0; i < n[thread_num]; i++) {

          double pp[3] = {pairs[thread_num][i].p1.x - centroid_m[thread_num][0],
            pairs[thread_num][i].p1.y - centroid_m[thread_num][1],
            pairs[thread_num][i].p1.z - centroid_m[thread_num][2]};
          double qq[3] = {pairs[thread_num][i].p2.x - centroid_d[thread_num][0],
            pairs[thread_num][i].p2.y - centroid_d[thread_num][1],
            pairs[thread_num][i].p2.z - centroid_d[thread_num][2]};
/*
          double pp[3] = {pairs[thread_num][i].p1.x - centroid_d[thread_num][0],
            pairs[thread_num][i].p1.y - centroid_d[thread_num][1],
            pairs[thread_num][i].p1.z - centroid_d[thread_num][2]};
          double qq[3] = {pairs[thread_num][i].p2.x - centroid_m[thread_num][0],
            pairs[thread_num][i].p2.y - centroid_m[thread_num][1],
            pairs[thread_num][i].p2.z - centroid_m[thread_num][2]};
*/
          // formula (6)
          Si[thread_num][0] += pp[0] * qq[0];
          Si[thread_num][1] += pp[0] * qq[1];
          Si[thread_num][2] += pp[0] * qq[2];
          Si[thread_num][3] += pp[1] * qq[0];
          Si[thread_num][4] += pp[1] * qq[1];
          Si[thread_num][5] += pp[1] * qq[2];
          Si[thread_num][6] += pp[2] * qq[0];
          Si[thread_num][7] += pp[2] * qq[1];
          Si[thread_num][8] += pp[2] * qq[2];
        }
      }
    } // end parallel
    
    // do we have enough point pairs?
    unsigned int pairssize = 0;
    for (int i = 0; i < OPENMP_NUM_THREADS; i++) {
      pairssize += n[i];
    }
    if (pairssize > 3) {
      if ((my_icp6Dminimizer->getAlgorithmID() == 1) ||
          (my_icp6Dminimizer->getAlgorithmID() == 2) ) {
        ret = my_icp6Dminimizer->Point_Point_Align_Parallel(OPENMP_NUM_THREADS,
            n, sum, centroid_m, centroid_d, Si, 
            alignxf);
      } else if (my_icp6Dminimizer->getAlgorithmID() == 6) {
        ret = my_icp6Dminimizer->Point_Point_Align_Parallel(OPENMP_NUM_THREADS,
            n, sum, centroid_m, centroid_d, 
            pairs,
            alignxf);
      } else {
        cout << "This parallel minimization algorithm is not implemented !!!" << endl;
        exit(-1);
      }
    } else {
      //break;
    }
#else

    double centroid_m[3] = {0.0, 0.0, 0.0};
    double centroid_d[3] = {0.0, 0.0, 0.0};
    vector<PtPair> pairs;
   
    Scan::getPtPairs(&pairs, PreviousScan, CurrentScan, 0, rnd,
        max_dist_match2, ret, centroid_m, centroid_d, pairing_mode);

    // do we have enough point pairs?
    if (pairs.size() > 3) {
      if (my_icp6Dminimizer->getAlgorithmID() == 3 || my_icp6Dminimizer->getAlgorithmID() == 8 ) {
        memcpy(alignxf, CurrentScan->get_transMat(), sizeof(alignxf));
      }
      ret = my_icp6Dminimizer->Point_Point_Align(pairs, alignxf, centroid_m, centroid_d);
    } else {
      break;
    }
   
#endif

    if ((iter == 0 && anim != -2) || ((anim > 0) && (iter % anim == 0))) {
      CurrentScan->transform(alignxf, Scan::ICP, 0);   // transform the current scan
    } else {
      CurrentScan->transform(alignxf, Scan::ICP, -1);  // transform the current scan
    }
    
    if ((fabs(ret - prev_ret) < epsilonICP) && (fabs(ret - prev_prev_ret) < epsilonICP)) {
      double id[16];
      M4identity(id);
      if(anim == -2) {
        CurrentScan->transform(id, Scan::ICP, -1);  // write end pose
      } else {
        CurrentScan->transform(id, Scan::ICP, 0);  // write end pose
      }
    break;
    }
  }
  
  return iter;
}


/**
 * Computes the point to point error between two scans 
 * 
 *
 */
double icp6D::Point_Point_Error(Scan* PreviousScan, Scan* CurrentScan, double max_dist_match, unsigned int *np) {
  double error = 0;
  unsigned int nr_ppairs = 0;

#ifdef _OPENMP
    omp_set_num_threads(OPENMP_NUM_THREADS);

    int max = (int)CurrentScan->size<DataXYZ>("xyz reduced");
    int step = max / OPENMP_NUM_THREADS;

    vector<PtPair> pairs[OPENMP_NUM_THREADS];
    double sum[OPENMP_NUM_THREADS];
    double centroid_m[OPENMP_NUM_THREADS][3];
    double centroid_d[OPENMP_NUM_THREADS][3];

    for (int i = 0; i < OPENMP_NUM_THREADS; i++) {
      sum[i] = centroid_m[i][0] = centroid_m[i][1] = centroid_m[i][2] = 0.0;
      centroid_d[i][0] = centroid_d[i][1] = centroid_d[i][2] = 0.0;
    }

#pragma omp parallel 
    {
      int thread_num = omp_get_thread_num();
      Scan::getPtPairsParallel(pairs, PreviousScan, CurrentScan,
          thread_num, step,
          rnd, sqr(max_dist_match),
          sum, centroid_m, centroid_d, CLOSEST_POINT);

    } 

    for (unsigned int thread_num = 0; thread_num < OPENMP_NUM_THREADS; thread_num++) {
      for (unsigned int i = 0; i < (unsigned int)pairs[thread_num].size(); i++) {
        error += sqr(pairs[thread_num][i].p1.x - pairs[thread_num][i].p2.x)
          + sqr(pairs[thread_num][i].p1.y - pairs[thread_num][i].p2.y)
          + sqr(pairs[thread_num][i].p1.z - pairs[thread_num][i].p2.z);
      }
      nr_ppairs += (unsigned int)pairs[thread_num].size();
    }
#else

    double centroid_m[3] = {0.0, 0.0, 0.0};
    double centroid_d[3] = {0.0, 0.0, 0.0};
    vector<PtPair> pairs;

    Scan::getPtPairs(&pairs, PreviousScan, CurrentScan, 0, rnd, sqr(max_dist_match),
                     error, centroid_m, centroid_d, CLOSEST_POINT);

    // getPtPairs computes error as sum of squared distances
    error = 0;

    for (unsigned int i = 0; i < pairs.size(); i++) {
      error += sqrt(
          sqr(pairs[i].p1.x - pairs[i].p2.x)
        + sqr(pairs[i].p1.y - pairs[i].p2.y)
        + sqr(pairs[i].p1.z - pairs[i].p2.z) );
    }
    nr_ppairs = pairs.size();
#endif

    if (np) *np = nr_ppairs;
//    return sqrt(error/nr_ppairs);
    return error/nr_ppairs;
}
/**
 * This function matches the scans only with ICP
 * 
 * @param allScans Contains all necessary scans.
 */
void icp6D::doICP(vector <Scan *> allScans, PairingMode pairing_mode)
{
  double id[16];
  M4identity(id);
  
  vector < Scan* > meta_scans;
  Scan* my_MetaScan = 0;

  
  for(unsigned int i = 0; i < allScans.size(); i++) {
    cout << i << "*" << endl;

    Scan *CurrentScan = allScans[i];
    Scan *PreviousScan = 0;
    
    if (i > 0) {
      PreviousScan = allScans[i-1];
      if (eP) {                             // extrapolate odometry
        CurrentScan->mergeCoordinatesWithRoboterPosition(PreviousScan);
      }
    }

    if (i > 0) {
      if (meta) {
        match(my_MetaScan, CurrentScan, pairing_mode);
      } else
      if (cad_matching) {
        match(allScans[0], CurrentScan, pairing_mode);
      } else {
        match(PreviousScan, CurrentScan, pairing_mode);
      }
    }

    // push processed scan
    if ( meta && i != allScans.size()-1 ) {
      meta_scans.push_back(CurrentScan);
      if (my_MetaScan) {
        delete my_MetaScan;
      }
      my_MetaScan = new MetaScan(meta_scans, nns_method, cuda_enabled);
    }
  }
}