ApolloPanIO.cpp

 
/* SPDX-License-Identifier: CC0-1.0 */
 
#include "ApolloPanIO.h"
 
#include <math.h>
#include <iostream>
 
#include "Ransac.h"
#include "stdio.h"
 
 
#define FIDL 5344.186 //nominal line spread of fiducials  in 5 micro pixels
#define FIDS 22980    //nominal sample spread of fiducials in 5 micron pixels
 
using namespace std;
 
namespace Isis {
 
  ApolloPanIO::ApolloPanIO() {
    this->initialize();
  }
 
  ApolloPanIO::~ApolloPanIO() {
    //empty destructor....
  }
 
  void ApolloPanIO::initialize() {
    int i;
 
    //the number of affines to be calculated
    n=0;
 
    //zero solved flags in discrete affines
    for (i=0; i<44; i++)
      affines[i].flag=0;
 
     //initial residuals statistic variable to nonsense values
     maxR  = -1.0;
     meanR = -1.0;
     stdevR= -1.0;
 
    //zero observed flags in fiducial observations
    for (i=0; i<90; i++) {
      obs[i].flag=0;
 
      //also calculate theoretical image observation of the fiducial marks
      if (i%2 == 0)
        obs[i].image[0] =  FIDS/2;
      else
        obs[i].image[0] = -FIDS/2;
 
      obs[i].image[1] = (-21.5 + double(int(i/2)))*FIDL;
 
      //adjustment to account for the half spacing among the 22nd, 23rd, and 24th
      //  fiducials on each side
      if (int(i/2)==22)
        obs[i].image[1] -= FIDL/2;
      else if ( int(i/2) > 22)
        obs[i].image[1] -= FIDL;
 
      obs[i].image[1] = -obs[i].image[1];  //sign reversal to match camera layout
    }
 
  }
 
  int ApolloPanIO::fiducialObservation(int fiducialNumber, double machineX, double machineY) {
    /*
    used to enter/overwrite observations of fiducial marks
 
    fiducial_number  -> number indicating which fiducial mark, see diagram at the top of the file
    machineX    -> x machine coordinate
    machineY    -> y machine coordinate
 
    returns 1 if the input data is reasonable otherwise -1
    */
 
 
    //first check the range of the fiducial number
    if (fiducialNumber < 0 || fiducialNumber > 89)
      return -1;
 
    //next check the range of the machine coordinates
    // (this is just a nonsense number check)
    if (fabs(machineX) > 1e20)
      return -1;
 
    if (fabs(machineY) > 1e20)
      return -1;
 
    //if all looks OK record the data
    obs[fiducialNumber].mach[0] = machineX;
    obs[fiducialNumber].mach[1] = machineY;
 
    //and mark the point observed
    obs[fiducialNumber].flag = 1;
 
    return 1;
  }
 
  int ApolloPanIO::clearFiducialObservation(int fiducialNumber) {
    //first check the range of the fiducial number
    if (fiducialNumber < 0 || fiducialNumber > 89)
      return -1;
 
    //and mark the point unobserved
    obs[fiducialNumber].flag = 0;
 
    return 1;
 
  }
 
  int ApolloPanIO::computeInteriorOrienation()
  {
    int     i,j,k,l,m;  //indeces-counters for linear algebra computations
 
    double  cdot[43][4][6],
        angle,              //angle of rotation to vertical for determining rotation coefficients
        det,                //determinate of 2x2 affine matrices used for matrix inversions
        //wcdot[43][4],    //all conditions are identically zero.. no need to store that..
        ndot[44][21],
        cndot[43][4][12],
        cxstar[172],     //cX*  matrix, Uotila course notes page 122
        cnct[14878],     //CNCt matrix, Uotila course notes page 122
        atw[44][6];      //right side of normal equation (see Uotila course notes)
                         // NOTE:  I store w with reversed signs compared to Uotila's notes
 
    double adot[2][6],wdot[2];
 
    //adot is a sub matrix of A (see Uotila Course notes page 122).
    //  It is composed of partials WRT to unknowns.
    //some affine partials are the same for every point so might as well hard code them now...
    adot[0][2] = 0.0;
    adot[0][3] = 0.0;
 
    adot[0][4] = 1.0;
    adot[0][5] = 0.0;
 
    adot[1][0] = 0.0;
    adot[1][1] = 0.0;
 
    adot[1][4] = 0.0;
    adot[1][5] = 1.0;
 
    //first order of business...
    // figure out how many affines there will be and which fiducial measures belong to each
 
    //the first affine always begins with point zero
    affines[0].indeces[0] =0;
    affines[0].flag = 1;
 
    for (i=2, n=0;  i<90;  i+=2) {
      // i is counting the fiducials
      // n is counting the affines
 
      //every time both fiducials in a horizontal (equal x coordinate) pair have been measured start
      //  a new affine
      if (obs[i].flag && obs[i+1].flag) {
        affines[n  ].indeces[1] = i+1;
        affines[n+1].indeces[0] = i  ;
        affines[n+1].flag=1;
        n++;
      }
    }
 
    //check for sufficient data
    if (n ==0) return -1;
 
    //now that we know how many affines will be needed we can get started
 
    //solve each affine individually
 
    //for each affine
    for (i =0; i<n; i++) {
      //initialize the appropriate ndot and atw matrices
      for (j=0; j<21;j++)
        ndot[i][j] =0.0;
 
      for (j=0; j<6; j++)
        atw[i][j]=0.0;
 
      //for each point pertaining to each affine
      for (j=affines[i].indeces[0]; j<=affines[i].indeces[1]; j++) {
        //if the point has been observed use the observation to build the normal
        if(obs[j].flag) {
          //store the partials that change...
          adot[0][0] = obs[j].mach[0];
          adot[0][1] = obs[j].mach[1];
 
          adot[1][2] = obs[j].mach[0];
          adot[1][3] = obs[j].mach[1];
 
          wdot[0] = obs[j].image[0];
          wdot[1] = obs[j].image[1];
 
 
          //build up the appropriate normal equations
          for (k=0; k<6; k++)
            for (l=0; l<=k; l++)
              for (m=0; m<2; m++)
                ndot[i][Isis::isymp(k,l)] += adot[m][k]*adot[m][l];
 
          for (k=0; k<6; k++)
            atw[i][k] += adot[0][k]*wdot[0] + adot[1][k]*wdot[1];
        }
      }
 
      //copy the atw matrix into the affine class structure so that it isn't lost when the solution
      // is done
      for (j=0; j<6; j++)
        affines[i].A2I[j] = atw[i][j];
 
      //solve this affine
      //NOTE: it must also be inverted for later calculations, hence three instead of two for the
      //  last argument
      if (!Isis::choleski_solve(ndot[i],affines[i].A2I,6,3))
        j=0;
    }
 
 
    //now the contributions of the continuity conditions must be added....
 
     //cdot[i] are sub matrices of C matrix (Uotila page 122)
     //Note: the cdot matrices are 4x12 (4 continuity conditions between 2 six paramter affines)
     //  however the left 4x6 portion of the cdot matrix and right 4x6 portion are equal but
     //  opposite so it's only necessary to calc/store one of them
    for (i=0; i<n-1; i++) {
      //build cdot matrix for continuity condition between affine i and i+1
      cdot[i][0][0 ] =  obs[affines[i].indeces[1]-1].mach[0];
      cdot[i][0][1 ] =  obs[affines[i].indeces[1]-1].mach[1];
      cdot[i][0][2 ] =  0.0;
      cdot[i][0][3 ] =  0.0;
      cdot[i][0][4 ] =  1.0;
      cdot[i][0][5 ] =  0.0;
 
      cdot[i][1][0 ] =  0.0;
      cdot[i][1][1 ] =  0.0;
      cdot[i][1][2 ] =  obs[affines[i].indeces[1]-1].mach[0];
      cdot[i][1][3 ] =  obs[affines[i].indeces[1]-1].mach[1];
      cdot[i][1][4 ] =  0.0;
      cdot[i][1][5 ] =  1.0;
 
      cdot[i][2][0 ] =  obs[affines[i].indeces[1]  ].mach[0];
      cdot[i][2][1 ] =  obs[affines[i].indeces[1]  ].mach[1];
      cdot[i][2][2 ] =  0.0;
      cdot[i][2][3 ] =  0.0;
      cdot[i][2][4 ] =  1.0;
      cdot[i][2][5 ] =  0.0;
 
      cdot[i][3][0 ] =  0.0;
      cdot[i][3][1 ] =  0.0;
      cdot[i][3][2 ] =  obs[affines[i].indeces[1]  ].mach[0];
      cdot[i][3][3 ] =  obs[affines[i].indeces[1]  ].mach[1];
      cdot[i][3][4 ] =  0.0;
      cdot[i][3][5 ] =  1.0;
    }
 
    //calculate cndot sub matrices of Uotila's C(AtM-A)- page122
    for (i=0; i<n-1; i++) {
      //initialize the matrix
      for (j=0; j<4; j++)
        for (k=0; k<12; k++)
          cndot[i][j][k] = 0.0;
 
      for (j=0; j<4; j++)
        for (k=0; k<6; k++)
          for (l=0; l<6; l++) {
            cndot[i][j][k  ] += cdot[i][j][l] * ndot[i  ][Isis::isymp(l,k)];
            cndot[i][j][k+6] -= cdot[i][j][l] * ndot[i+1][Isis::isymp(l,k)];
          }
    }
 
    //calculate the CX* matrix -page 122
    for (i=0; i<(n-1)*4; i++) {
      j= int(i/4);  //which c sub matrix
      k= i%4;      //row of the c sub matrix
 
      cxstar[i] = 0;
 
      for (l=0; l<6; l++) {
        cxstar[i] -= cdot[j][k][l]*affines[j  ].A2I[l];
        cxstar[i] += cdot[j][k][l]*affines[j+1].A2I[l];
      }
    }
 
     //initialize CNCt matrix
     for (i=0; i<14878; i++)
        cnct[i] = 0.0;
 
    //calculate all of the diagonal portions
    for (i=0; i<n-1; i++) {
        for (j=0; j<4; j++)
           for (k=0; k<=j; k++) //remember that this matrix is a memory optimized symmetric matrix
              for (l=0; l<12; l++) {
                 if(l<6)
                    cnct[Isis::isymp(j+4*i,k+4*i)] += cndot[i][j][l] * cdot[i][k][l];
                 else
                    cnct[Isis::isymp(j+4*i,k+4*i)] -= cndot[i][j][l] * cdot[i][k][l-6];
              }
    }
 
     //calculate all of the diagonal portions
     for (i=0; i<n-2; i++) {
        for (j=0; j<4; j++)
           for (k=0; k<4; k++)
              for (l=0; l<6; l++)
                 cnct[Isis::isymp(j+i*4,k+4+i*4)] += cndot[i][j][l+6]*cdot[i+1][k][l];
     }
 
 
     Isis::choleski_solve(cnct,cxstar,4*(n-1),2);
 
     //cxstar are now Kc Lagrange multipliers
     //now (CN)tKc = NCtKc are the secondary corrections to the unknowns and they can be added
     //  directly to the previously calculated values
 
     //corrections to the first affine
     for (i=0; i<6; i++) {
        for (j=0; j<4;j ++)
           affines[0].A2I[i] += cxstar[j]*cndot[0][j][i];
     }
 
     //for all the middle affines
     for (i=1; i<n-1; i++) {
        for (j=0; j<6; j++) {
           for (k=0; k<4; k++)
              affines[i].A2I[j] += cxstar[(i-1)*4+k]*cndot[i-1][k][j+6];
           for (k=0; k<4; k++)
              affines[i].A2I[j] += cxstar[(i  )*4+k]*cndot[i  ][k][j  ];
        }
     }
 
     //corrections for the last affine
     for (i=0; i<6; i++) {
        for (j=0; j<4; j++)
           affines[n-1].A2I[i] += cxstar[4*(n-2)+j]*cndot[n-2][j][i+6];
     }
 
 
     //now compute the reverse affines
     for (i=0; i<n; i++) {
        det = affines[i].A2I[0]*affines[i].A2I[3] - affines[i].A2I[1]*affines[i].A2I[2];
 
        affines[i].A2M[0] =  affines[i].A2I[3]/det;
        affines[i].A2M[3] =  affines[i].A2I[0]/det;
 
        affines[i].A2M[1] = -affines[i].A2I[1]/det;
        affines[i].A2M[2] = -affines[i].A2I[2]/det;
 
        affines[i].A2M[4] = - affines[i].A2M[0]*affines[i].A2I[4]
                            - affines[i].A2M[1]*affines[i].A2I[5];
        affines[i].A2M[5] = - affines[i].A2M[2]*affines[i].A2I[4]
                            - affines[i].A2M[3]*affines[i].A2I[5];
     }
 
     //now define the rotation coefficients that will transform any image/machine y coordinate into
     //  a reference frame where the two right most fiducials are have the exact same y value and
     //  it is thus easy to determine where a point lies on the left side of the line
 
     double pt1[2], pt2[2],v[2];
 
     for (i=0; i<n; i++) {
        //rotations in machine space,
 
        //the two points obs[affines[i].indeces[1]-1] and obs[affines[i].indeces[1]] define a line
 
        pt1[0] = obs[affines[i].indeces[1]-1].mach[0];
        pt1[1] = obs[affines[i].indeces[1]-1].mach[1];
 
        pt2[0] = obs[affines[i].indeces[1]  ].mach[0];
        pt2[1] = obs[affines[i].indeces[1]  ].mach[1];
 
 
        //find the point on the line closest to the origin
 
        //vector in the direction of a line
        v[0] = pt1[0] - pt2[0];
        v[1] = pt1[1] - pt2[1];
 
        //using det as the scale factor along the line from pt1 to the point nearest the origin
        det = -(pt1[0]*v[0] + pt1[1]*v[1])/(v[0]*v[0]+v[1]*v[1]);
 
        //pt1 + det*v is the point closest to the origin and the angle needed to rotate that
        //  point so that it lies on the y axis is...
        angle = atan2((pt1[0]+det*v[0]),(pt1[1]+det*v[1]));
 
        //and since we're only ever going to be concerned with rotating the y coordinates and z's
        //  are alwyas zero there are only two values we need to save to elements of the rotation
        //  matrix
        affines[i].rotM[0] =  sin(angle);
        affines[i].rotM[1] =  cos(angle);
 
        //any machine coordinate [u,v] can be rotated into a system where the greatest y values of
        // the fiducial marks are the same, and therefore the greatest y value in a particular
        // affine is:
        affines[i].mM = pt1[0]*affines[i].rotM[0] + pt1[1]*affines[i].rotM[1];
 
        //or equivalently...
        //affines[i].mM = pt2[0]*affines[i].rotM[0] + pt2[1]*affines[i].rotM[1];
 
 
 
        //rotations in image space
 
        //for this calculation the transformed Machine (measured coordinates) are used rather than
        //  the theoretical/expected coordinates of the fiducial marks.  They should be close to
        //  identical, but it is the affines and therefore the transformed points that define the
        //  actual work space.
        pt1[0] = affines[i].A2I[0] * obs[affines[i].indeces[1]-1].mach[0] +
                 affines[i].A2I[1] * obs[affines[i].indeces[1]-1].mach[1] +
                 affines[i].A2I[4];
 
        pt1[1] = affines[i].A2I[2] * obs[affines[i].indeces[1]-1].mach[0] +
                 affines[i].A2I[3] * obs[affines[i].indeces[1]-1].mach[1] +
                 affines[i].A2I[5];
 
        pt2[0] = affines[i].A2I[0] * obs[affines[i].indeces[1]  ].mach[0] +
                 affines[i].A2I[1] * obs[affines[i].indeces[1]  ].mach[1] +
                 affines[i].A2I[4];
 
        pt2[1] = affines[i].A2I[2] * obs[affines[i].indeces[1]  ].mach[0] +
                 affines[i].A2I[3] * obs[affines[i].indeces[1]  ].mach[1] +
                 affines[i].A2I[5];
 
 
        //find the point on the line closest to the origin
 
        //vector in the direction of a line
        v[0] = pt1[0] - pt2[0];
        v[1] = pt1[1] - pt2[1];
 
        //using det as the scale factor along the line from pt1 to the point nearest the origin
        det = -(pt1[0]*v[0] + pt1[1]*v[1])/(v[0]*v[0]+v[1]*v[1]);
 
        //pt1 + det*v is the point closest to the origin and the angle needed to rotate that point
        // so that it lies on the y axis is...
        angle = atan((pt1[0]+det*v[0])/(pt1[1]+det*v[1]));
 
        //and since we're only ever going to be concerned with rotating the y coordinates and z's
        // are alwyas zero there are only two values we need to save to elements of the rotation
        // matrix
        affines[i].rotI[0] =  sin(angle);
        affines[i].rotI[1] =  cos(angle);
 
        //any machine coordinate [u,v] can be rotated into a system where greatest y values of the
        // affine border are the same that is the border is a line parrallel to the image x axis,
        // and therefore the greatest y value in a particular affine is
        affines[i].mI = pt1[0]*affines[i].rotI[0] + pt1[1]*affines[i].rotI[1];
 
        //or equivalently...
     }
 
 
 
 
 
 
     //all the parameters necessary for the machine2image and image2machin coordinate
     //  transformations are now computed and stored
 
 
 
 
 
 
     //calculate and store residuals
     for (i=0; i<90; i++) {
        if (obs[i].flag) {   //if the point has been observed...
           //convert the theoretical image coordinate to machine space and compare it to the
           // measured coordinate--this way residuals are in pixels
           image2Machine(obs[i].image[0],obs[i].image[1],&pt1[0],&pt1[1]);
 
           obs[i].residuals[0] = pt1[0] - obs[i].mach[0];
           obs[i].residuals[1] = pt1[1] - obs[i].mach[1];
        }
     }
 
     calc_residual_stats();  //DEBUG
 
     return 1;
  }
 
  int ApolloPanIO::machine2Image(double *machineX, double *machineY) {
 
    double temp,npt[2];
    int i;
 
    //first do a domain check to be sure this point is in the last affine
    temp = *machineX * affines[n-2].rotM[0] + *machineY * affines[n-2].rotM[1];
 
    if (temp > affines[n-2].mM) {
      i=n-1;
    }
    else {//now step through the affines to determine in the domain of which one this point lies
      for (i=0; i<n-1; i++) {
        temp = *machineX * affines[i].rotM[0] + *machineY * affines[i].rotM[1];
 
        //if the point is in the domain of this affine do the conversion...
        if (temp <= affines[i].mM)
          break;
      }
    }
    //cout << "DEBUG: affine index: " << i << endl;
    //cout << "affine " << affines[i].A2I[0] << " " << affines[i].A2I[1] << " "<< affines[i].A2I[2] << " "<< affines[i].A2I[3] << " "<< affines[i].A2I[4] << " "<< affines[i].A2I[5];
    npt[0] = affines[i].A2I[0] * *machineX + affines[i].A2I[1] * *machineY + affines[i].A2I[4];
    npt[1] = affines[i].A2I[2] * *machineX + affines[i].A2I[3] * *machineY + affines[i].A2I[5];
 
    *machineX = npt[0];  //overwrite the input
    *machineY = npt[1];  //overwrite the input
 
    return 1;    //point successfully converted
  }
 
  int ApolloPanIO::machine2Image(double machineX,
                                 double machineY,
                                 double *imageX,
                                 double *imageY) {
 
     int i;
 
     *imageX = machineX;
     *imageY = machineY;
 
     i = machine2Image(imageX,imageY);
 
     return i;
  }
 
  int ApolloPanIO::image2Machine(double *imageX, double *imageY) {
 
    double temp, npt[2];
    int i;
 
    //first do a domain check if this point is in the last affine
    temp = *imageX * affines[n-2].rotI[0] + *imageY * affines[n-2].rotI[1];
 
    if( temp < affines[n-2].mI)
      i = n-1;
    else {
      //now step through the affines to determine in the domain of which one this point lies
      for (i=0; i<n-1; i++) {
        temp = *imageX * affines[i].rotI[0] + *imageY * affines[i].rotI[1];
 
        //if the point is in the domain of this affine do the conversion...
        if( temp >= affines[i].mI)
          break;
      }
    }
    //cout << "DEBUG: affine index: " << i << endl;
    //cout << "affine " << affines[i].A2M[0] << " " << affines[i].A2M[1] << " "<< affines[i].A2M[2] << " "<< affines[i].A2M[3] << " "<< affines[i].A2M[4] << " "<< affines[i].A2M[5];
    npt[0] = affines[i].A2M[0] * *imageX + affines[i].A2M[1] * *imageY + affines[i].A2M[4];
    npt[1] = affines[i].A2M[2] * *imageX + affines[i].A2M[3] * *imageY + affines[i].A2M[5];
 
    *imageX = npt[0];  //overwrite the input
    *imageY = npt[1];  //overwrite the input
 
    return 1;    //point successfully converted
  }
 
  int ApolloPanIO::image2Machine(double imageX,
                                 double imageY,
                                 double *machineX,
                                 double *machineY) {
     /* This class method converts a machine coordinate to an image coordinate
        (overwriting the input arguments)
 
        imageX -> machine x input
        imageY -> machine y input
        machineX   <- to be overwritten as the converted coordinate
        machineY   <- to be overwritten as the converted coordinate
     */
 
     *machineX = imageX;
     *machineY = imageY;
 
     return image2Machine(machineX,machineY);
  }
 
  void ApolloPanIO::calc_residual_stats() {
      maxR = 0.0;
 
      meanR = 0.0;
 
      stdevR = 0.0;
 
      int i,n=0;
 
      double l;
 
      for (i=0; i<90; i++) {
        if (obs[i].flag) {
           l = sqrt(obs[i].residuals[0]*obs[i].residuals[0] +
                    obs[i].residuals[1]*obs[i].residuals[1]);
 
           if( l > maxR)
              maxR = l;
 
           meanR += l;
 
           n++;
        }
      }
 
      meanR /=n;
 
      for (i=0; i<90; i++) {
         if(obs[i].flag) {
            l = sqrt(obs[i].residuals[0]*obs[i].residuals[0] +
                     obs[i].residuals[1]*obs[i].residuals[1]);
 
            stdevR += (l-meanR)*(l-meanR);
         }
      }
 
      stdevR = sqrt(stdevR/(n-1));
 
     return;
  }
 
  double ApolloPanIO::stdevResiduals() {
    //returns the standard deviation of the length of the 2D residual vectors
       //if it hasn't been calucated a nonsense number (-1.0) will be returned
    return stdevR;
  }
 
  double ApolloPanIO::meanResiduals()
  {
    return meanR;
  }
 
  double ApolloPanIO::maxResiduals()
  {
    return maxR;
  }
} //end namespace Isis