← Back to branch summary

~ubuntu-branches/ubuntu/trusty/hugin/trusty-proposed

« back to all changes in this revision

Viewing changes to src/hugin_cpfind/localfeatures/RansacFiltering.cpp

  • Committer: Bazaar Package Importer
  • Author(s): Andreas Metzler
  • Date: 2011-01-06 14:28:24 UTC
  • mfrom: (1.1.9 upstream) (0.1.21 experimental)
  • Revision ID: james.westby@ubuntu.com-20110106142824-zn9lxylg5z44dynn
* Drop Cyril Brulebois from Uploaders. Thank you very much for your work.
* Bump package version. (rc3 was re-released as 2010.4.0).

Show diffs side-by-side

added added

removed removed

Lines of Context:
src/hugin_cpfind/localfeatures/RansacFiltering.cpp
 
1
/*
 
2
* Copyright (C) 2007-2008 Anael Orlinski
 
3
*
 
4
* This file is part of Panomatic.
 
5
*
 
6
* Panomatic is free software; you can redistribute it and/or modify
 
7
* it under the terms of the GNU General Public License as published by
 
8
* the Free Software Foundation; either version 2 of the License, or
 
9
* (at your option) any later version.
 
10
* 
 
11
* Panomatic is distributed in the hope that it will be useful,
 
12
* but WITHOUT ANY WARRANTY; without even the implied warranty of
 
13
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 
14
* GNU General Public License for more details.
 
15
* 
 
16
* You should have received a copy of the GNU General Public License
 
17
* along with Panomatic; if not, write to the Free Software
 
18
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 
19
*/
 
20
 
 
21
#include "RansacFiltering.h"
 
22
#include "Homography.h"
 
23
 
 
24
#include <boost/foreach.hpp>
 
25
 
 
26
using namespace std;
 
27
 
 
28
static int genint(int x)
 
29
{
 
30
        return (int)((double)rand()*x/(double)RAND_MAX);
 
31
}
 
32
 
 
33
namespace lfeat
 
34
{
 
35
 
 
36
// distance between estimate and real point in pixels.
 
37
 
 
38
double Ransac::calcError(Homography *aH, PointMatch &aM)
 
39
{
 
40
        double x1p, y1p;
 
41
        aH->transformPoint(aM._img1_x, aM._img1_y, x1p, y1p);
 
42
        
 
43
        double d1 = aM._img2_x - x1p;
 
44
        double d2 = aM._img2_y - y1p;
 
45
 
 
46
        return d1*d1+d2*d2;
 
47
}
 
48
 
 
49
 
 
50
void Ransac::filter(PointMatchVector_t& ioMatches, PointMatchVector_t& ioRemovedMatches)
 
51
{
 
52
        int aRemainingIterations = _nIter;
 
53
        const double aErrorDistSq = _distanceThres * _distanceThres;
 
54
        
 
55
        Homography aCurrentModel;
 
56
        
 
57
        unsigned int aMaxInliers = 0;
 
58
        PointMatchVector_t aBestInliers;
 
59
        PointMatchVector_t aBestOutliers;
 
60
 
 
61
        // normalization  !!!!!!
 
62
        aCurrentModel.initMatchesNormalization(ioMatches);
 
63
 
 
64
 
 
65
        //std::cout << "gravity " << _v1x << " " << _v1y << " " << _v2x << " " << _v2y << endl;
 
66
 
 
67
        for(; aRemainingIterations > 0; aRemainingIterations--)
 
68
        {
 
69
                //cout << "RANSAC -- iter " << aRemainingIterations << endl;
 
70
                
 
71
                // random select 4 matches to fit the model
 
72
                // from the input set as maybe_inliers
 
73
                PointMatchVector_t aMatches(ioMatches);
 
74
                PointMatchVector_t aInliers, aOutliers;
 
75
 
 
76
                //std::cout << aMatches.size() << " matches size" << endl;
 
77
 
 
78
                for(int i=0; i<5; i++)
 
79
                {
 
80
                        int n = genint((int)aMatches.size()-1);
 
81
                        aInliers.push_back(aMatches.at(n));
 
82
                        aMatches.erase(aMatches.begin()+n);
 
83
                }
 
84
 
 
85
                //std::cout << aMatches.size() << " estimate" << endl;
 
86
 
 
87
                if (!aCurrentModel.estimate(aInliers))
 
88
                {
 
89
                        aMatches.clear();
 
90
                        continue;
 
91
                }
 
92
 
 
93
                // for every point remaining in aMatches, add them to aInliers if they fit the model well. 
 
94
                BOOST_FOREACH(PointMatchPtr aMatchesIter2, aMatches)
 
95
                {
 
96
                        if (calcError(&aCurrentModel, *aMatchesIter2) < aErrorDistSq)
 
97
                        {
 
98
                                aInliers.push_back(aMatchesIter2);
 
99
                        }
 
100
                        else
 
101
                        {
 
102
                                aOutliers.push_back(aMatchesIter2);
 
103
                        }
 
104
                }
 
105
                
 
106
                if (aInliers.size() > aMaxInliers)
 
107
                {
 
108
                        //cout << "good found -----------------" << aRemainingIterations << endl;
 
109
                        //cout << aCurrentModel << endl;
 
110
                        for (int i=0; i<3;++i)
 
111
                                for(int j=0; j<3;++j)
 
112
                                        _bestModel._H[i][j] = aCurrentModel._H[i][j];
 
113
                        
 
114
                        _bestModel._v1x = aCurrentModel._v1x;
 
115
                        _bestModel._v2x = aCurrentModel._v2x;
 
116
                        _bestModel._v1y = aCurrentModel._v1y;
 
117
                        _bestModel._v2y = aCurrentModel._v2y;
 
118
                        
 
119
                        
 
120
                        //*ioBestModel = *aCurrentModel;
 
121
                        aMaxInliers = (unsigned int)aInliers.size();
 
122
                        aBestInliers = aInliers;
 
123
                        aBestOutliers = aOutliers;
 
124
                        //cout << "Inliers : " << aInliers.size() << " Outliers : " << aOutliers.size() << endl;
 
125
 
 
126
                }
 
127
 
 
128
                // if there are 0 outliers then we are done.
 
129
                if (aOutliers.empty())
 
130
                        break;
 
131
 
 
132
                //cout << "Inliers : " << aInliers.size() << " Outliers : " << aOutliers.size() << endl;
 
133
                //cout << "end iter" << endl;
 
134
        }
 
135
 
 
136
        ioMatches = aBestInliers;
 
137
        ioRemovedMatches = aBestOutliers;
 
138
 
 
139
 
 
140
}
 
141
 
 
142
void Ransac::transform(double iX, double iY, double &oX, double &oY)
 
143
{
 
144
        _bestModel.transformPoint(iX, iY, oX, oY);
 
145
}
 
146
 
 
147
}