1 /*M///////////////////////////////////////////////////////////////////////////////////////
2 //
3 // IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
4 //
5 // By downloading, copying, installing or using the software you agree to this license.
6 // If you do not agree to this license, do not download, install,
7 // copy or use the software.
8 //
9 //
10 // License Agreement
11 // For Open Source Computer Vision Library
12 //
13 // Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
14 // Copyright (C) 2009-2011, Willow Garage Inc., all rights reserved.
15 // Third party copyrights are property of their respective owners.
16 //
17 // Redistribution and use in source and binary forms, with or without modification,
18 // are permitted provided that the following conditions are met:
19 //
20 // * Redistribution's of source code must retain the above copyright notice,
21 // this list of conditions and the following disclaimer.
22 //
23 // * Redistribution's in binary form must reproduce the above copyright notice,
24 // this list of conditions and the following disclaimer in the documentation
25 // and/or other materials provided with the distribution.
26 //
27 // * The name of the copyright holders may not be used to endorse or promote products
28 // derived from this software without specific prior written permission.
29 //
30 // This software is provided by the copyright holders and contributors "as is" and
31 // any express or implied warranties, including, but not limited to, the implied
32 // warranties of merchantability and fitness for a particular purpose are disclaimed.
33 // In no event shall the Intel Corporation or contributors be liable for any direct,
34 // indirect, incidental, special, exemplary, or consequential damages
35 // (including, but not limited to, procurement of substitute goods or services;
36 // loss of use, data, or profits; or business interruption) however caused
37 // and on any theory of liability, whether in contract, strict liability,
38 // or tort (including negligence or otherwise) arising in any way out of
39 // the use of this software, even if advised of the possibility of such damage.
40 //
41 //M*/
42
43 #include "precomp.hpp"
44 #include "fisheye.hpp"
45 #include <limits>
46
47 namespace cv { namespace
48 {
49 struct JacobianRow
50 {
51 Vec2d df, dc;
52 Vec4d dk;
53 Vec3d dom, dT;
54 double dalpha;
55 };
56
57 void subMatrix(const Mat& src, Mat& dst, const std::vector<uchar>& cols, const std::vector<uchar>& rows);
58 }}
59
60 //////////////////////////////////////////////////////////////////////////////////////////////////////////////
61 /// cv::fisheye::projectPoints
62
projectPoints(InputArray objectPoints,OutputArray imagePoints,const Affine3d & affine,InputArray K,InputArray D,double alpha,OutputArray jacobian)63 void cv::fisheye::projectPoints(InputArray objectPoints, OutputArray imagePoints, const Affine3d& affine,
64 InputArray K, InputArray D, double alpha, OutputArray jacobian)
65 {
66 CV_INSTRUMENT_REGION();
67
68 projectPoints(objectPoints, imagePoints, affine.rvec(), affine.translation(), K, D, alpha, jacobian);
69 }
70
projectPoints(InputArray objectPoints,OutputArray imagePoints,InputArray _rvec,InputArray _tvec,InputArray _K,InputArray _D,double alpha,OutputArray jacobian)71 void cv::fisheye::projectPoints(InputArray objectPoints, OutputArray imagePoints, InputArray _rvec,
72 InputArray _tvec, InputArray _K, InputArray _D, double alpha, OutputArray jacobian)
73 {
74 CV_INSTRUMENT_REGION();
75
76 // will support only 3-channel data now for points
77 CV_Assert(objectPoints.type() == CV_32FC3 || objectPoints.type() == CV_64FC3);
78 imagePoints.create(objectPoints.size(), CV_MAKETYPE(objectPoints.depth(), 2));
79 size_t n = objectPoints.total();
80
81 CV_Assert(_rvec.total() * _rvec.channels() == 3 && (_rvec.depth() == CV_32F || _rvec.depth() == CV_64F));
82 CV_Assert(_tvec.total() * _tvec.channels() == 3 && (_tvec.depth() == CV_32F || _tvec.depth() == CV_64F));
83 CV_Assert(_tvec.getMat().isContinuous() && _rvec.getMat().isContinuous());
84
85 Vec3d om = _rvec.depth() == CV_32F ? (Vec3d)*_rvec.getMat().ptr<Vec3f>() : *_rvec.getMat().ptr<Vec3d>();
86 Vec3d T = _tvec.depth() == CV_32F ? (Vec3d)*_tvec.getMat().ptr<Vec3f>() : *_tvec.getMat().ptr<Vec3d>();
87
88 CV_Assert(_K.size() == Size(3,3) && (_K.type() == CV_32F || _K.type() == CV_64F) && _D.type() == _K.type() && _D.total() == 4);
89
90 cv::Vec2d f, c;
91 if (_K.depth() == CV_32F)
92 {
93
94 Matx33f K = _K.getMat();
95 f = Vec2f(K(0, 0), K(1, 1));
96 c = Vec2f(K(0, 2), K(1, 2));
97 }
98 else
99 {
100 Matx33d K = _K.getMat();
101 f = Vec2d(K(0, 0), K(1, 1));
102 c = Vec2d(K(0, 2), K(1, 2));
103 }
104
105 Vec4d k = _D.depth() == CV_32F ? (Vec4d)*_D.getMat().ptr<Vec4f>(): *_D.getMat().ptr<Vec4d>();
106
107 const bool isJacobianNeeded = jacobian.needed();
108 JacobianRow *Jn = 0;
109 if (isJacobianNeeded)
110 {
111 int nvars = 2 + 2 + 1 + 4 + 3 + 3; // f, c, alpha, k, om, T,
112 jacobian.create(2*(int)n, nvars, CV_64F);
113 Jn = jacobian.getMat().ptr<JacobianRow>(0);
114 }
115
116 Matx33d R;
117 Matx<double, 3, 9> dRdom;
118 Rodrigues(om, R, dRdom);
119 Affine3d aff(om, T);
120
121 const Vec3f* Xf = objectPoints.getMat().ptr<Vec3f>();
122 const Vec3d* Xd = objectPoints.getMat().ptr<Vec3d>();
123 Vec2f *xpf = imagePoints.getMat().ptr<Vec2f>();
124 Vec2d *xpd = imagePoints.getMat().ptr<Vec2d>();
125
126 for(size_t i = 0; i < n; ++i)
127 {
128 Vec3d Xi = objectPoints.depth() == CV_32F ? (Vec3d)Xf[i] : Xd[i];
129 Vec3d Y = aff*Xi;
130 if (fabs(Y[2]) < DBL_MIN)
131 Y[2] = 1;
132 Vec2d x(Y[0]/Y[2], Y[1]/Y[2]);
133
134 double r2 = x.dot(x);
135 double r = std::sqrt(r2);
136
137 // Angle of the incoming ray:
138 double theta = atan(r);
139
140 double theta2 = theta*theta, theta3 = theta2*theta, theta4 = theta2*theta2, theta5 = theta4*theta,
141 theta6 = theta3*theta3, theta7 = theta6*theta, theta8 = theta4*theta4, theta9 = theta8*theta;
142
143 double theta_d = theta + k[0]*theta3 + k[1]*theta5 + k[2]*theta7 + k[3]*theta9;
144
145 double inv_r = r > 1e-8 ? 1.0/r : 1;
146 double cdist = r > 1e-8 ? theta_d * inv_r : 1;
147
148 Vec2d xd1 = x * cdist;
149 Vec2d xd3(xd1[0] + alpha*xd1[1], xd1[1]);
150 Vec2d final_point(xd3[0] * f[0] + c[0], xd3[1] * f[1] + c[1]);
151
152 if (objectPoints.depth() == CV_32F)
153 xpf[i] = final_point;
154 else
155 xpd[i] = final_point;
156
157 if (isJacobianNeeded)
158 {
159 //Vec3d Xi = pdepth == CV_32F ? (Vec3d)Xf[i] : Xd[i];
160 //Vec3d Y = aff*Xi;
161 double dYdR[] = { Xi[0], Xi[1], Xi[2], 0, 0, 0, 0, 0, 0,
162 0, 0, 0, Xi[0], Xi[1], Xi[2], 0, 0, 0,
163 0, 0, 0, 0, 0, 0, Xi[0], Xi[1], Xi[2] };
164
165 Matx33d dYdom_data = Matx<double, 3, 9>(dYdR) * dRdom.t();
166 const Vec3d *dYdom = (Vec3d*)dYdom_data.val;
167
168 Matx33d dYdT_data = Matx33d::eye();
169 const Vec3d *dYdT = (Vec3d*)dYdT_data.val;
170
171 //Vec2d x(Y[0]/Y[2], Y[1]/Y[2]);
172 Vec3d dxdom[2];
173 dxdom[0] = (1.0/Y[2]) * dYdom[0] - x[0]/Y[2] * dYdom[2];
174 dxdom[1] = (1.0/Y[2]) * dYdom[1] - x[1]/Y[2] * dYdom[2];
175
176 Vec3d dxdT[2];
177 dxdT[0] = (1.0/Y[2]) * dYdT[0] - x[0]/Y[2] * dYdT[2];
178 dxdT[1] = (1.0/Y[2]) * dYdT[1] - x[1]/Y[2] * dYdT[2];
179
180 //double r2 = x.dot(x);
181 Vec3d dr2dom = 2 * x[0] * dxdom[0] + 2 * x[1] * dxdom[1];
182 Vec3d dr2dT = 2 * x[0] * dxdT[0] + 2 * x[1] * dxdT[1];
183
184 //double r = std::sqrt(r2);
185 double drdr2 = r > 1e-8 ? 1.0/(2*r) : 1;
186 Vec3d drdom = drdr2 * dr2dom;
187 Vec3d drdT = drdr2 * dr2dT;
188
189 // Angle of the incoming ray:
190 //double theta = atan(r);
191 double dthetadr = 1.0/(1+r2);
192 Vec3d dthetadom = dthetadr * drdom;
193 Vec3d dthetadT = dthetadr * drdT;
194
195 //double theta_d = theta + k[0]*theta3 + k[1]*theta5 + k[2]*theta7 + k[3]*theta9;
196 double dtheta_ddtheta = 1 + 3*k[0]*theta2 + 5*k[1]*theta4 + 7*k[2]*theta6 + 9*k[3]*theta8;
197 Vec3d dtheta_ddom = dtheta_ddtheta * dthetadom;
198 Vec3d dtheta_ddT = dtheta_ddtheta * dthetadT;
199 Vec4d dtheta_ddk = Vec4d(theta3, theta5, theta7, theta9);
200
201 //double inv_r = r > 1e-8 ? 1.0/r : 1;
202 //double cdist = r > 1e-8 ? theta_d / r : 1;
203 Vec3d dcdistdom = inv_r * (dtheta_ddom - cdist*drdom);
204 Vec3d dcdistdT = inv_r * (dtheta_ddT - cdist*drdT);
205 Vec4d dcdistdk = inv_r * dtheta_ddk;
206
207 //Vec2d xd1 = x * cdist;
208 Vec4d dxd1dk[2];
209 Vec3d dxd1dom[2], dxd1dT[2];
210 dxd1dom[0] = x[0] * dcdistdom + cdist * dxdom[0];
211 dxd1dom[1] = x[1] * dcdistdom + cdist * dxdom[1];
212 dxd1dT[0] = x[0] * dcdistdT + cdist * dxdT[0];
213 dxd1dT[1] = x[1] * dcdistdT + cdist * dxdT[1];
214 dxd1dk[0] = x[0] * dcdistdk;
215 dxd1dk[1] = x[1] * dcdistdk;
216
217 //Vec2d xd3(xd1[0] + alpha*xd1[1], xd1[1]);
218 Vec4d dxd3dk[2];
219 Vec3d dxd3dom[2], dxd3dT[2];
220 dxd3dom[0] = dxd1dom[0] + alpha * dxd1dom[1];
221 dxd3dom[1] = dxd1dom[1];
222 dxd3dT[0] = dxd1dT[0] + alpha * dxd1dT[1];
223 dxd3dT[1] = dxd1dT[1];
224 dxd3dk[0] = dxd1dk[0] + alpha * dxd1dk[1];
225 dxd3dk[1] = dxd1dk[1];
226
227 Vec2d dxd3dalpha(xd1[1], 0);
228
229 //final jacobian
230 Jn[0].dom = f[0] * dxd3dom[0];
231 Jn[1].dom = f[1] * dxd3dom[1];
232
233 Jn[0].dT = f[0] * dxd3dT[0];
234 Jn[1].dT = f[1] * dxd3dT[1];
235
236 Jn[0].dk = f[0] * dxd3dk[0];
237 Jn[1].dk = f[1] * dxd3dk[1];
238
239 Jn[0].dalpha = f[0] * dxd3dalpha[0];
240 Jn[1].dalpha = 0; //f[1] * dxd3dalpha[1];
241
242 Jn[0].df = Vec2d(xd3[0], 0);
243 Jn[1].df = Vec2d(0, xd3[1]);
244
245 Jn[0].dc = Vec2d(1, 0);
246 Jn[1].dc = Vec2d(0, 1);
247
248 //step to jacobian rows for next point
249 Jn += 2;
250 }
251 }
252 }
253
254 //////////////////////////////////////////////////////////////////////////////////////////////////////////////
255 /// cv::fisheye::distortPoints
256
distortPoints(InputArray undistorted,OutputArray distorted,InputArray K,InputArray D,double alpha)257 void cv::fisheye::distortPoints(InputArray undistorted, OutputArray distorted, InputArray K, InputArray D, double alpha)
258 {
259 CV_INSTRUMENT_REGION();
260
261 // will support only 2-channel data now for points
262 CV_Assert(undistorted.type() == CV_32FC2 || undistorted.type() == CV_64FC2);
263 distorted.create(undistorted.size(), undistorted.type());
264 size_t n = undistorted.total();
265
266 CV_Assert(K.size() == Size(3,3) && (K.type() == CV_32F || K.type() == CV_64F) && D.total() == 4);
267
268 cv::Vec2d f, c;
269 if (K.depth() == CV_32F)
270 {
271 Matx33f camMat = K.getMat();
272 f = Vec2f(camMat(0, 0), camMat(1, 1));
273 c = Vec2f(camMat(0, 2), camMat(1, 2));
274 }
275 else
276 {
277 Matx33d camMat = K.getMat();
278 f = Vec2d(camMat(0, 0), camMat(1, 1));
279 c = Vec2d(camMat(0 ,2), camMat(1, 2));
280 }
281
282 Vec4d k = D.depth() == CV_32F ? (Vec4d)*D.getMat().ptr<Vec4f>(): *D.getMat().ptr<Vec4d>();
283
284 const Vec2f* Xf = undistorted.getMat().ptr<Vec2f>();
285 const Vec2d* Xd = undistorted.getMat().ptr<Vec2d>();
286 Vec2f *xpf = distorted.getMat().ptr<Vec2f>();
287 Vec2d *xpd = distorted.getMat().ptr<Vec2d>();
288
289 for(size_t i = 0; i < n; ++i)
290 {
291 Vec2d x = undistorted.depth() == CV_32F ? (Vec2d)Xf[i] : Xd[i];
292
293 double r2 = x.dot(x);
294 double r = std::sqrt(r2);
295
296 // Angle of the incoming ray:
297 double theta = atan(r);
298
299 double theta2 = theta*theta, theta3 = theta2*theta, theta4 = theta2*theta2, theta5 = theta4*theta,
300 theta6 = theta3*theta3, theta7 = theta6*theta, theta8 = theta4*theta4, theta9 = theta8*theta;
301
302 double theta_d = theta + k[0]*theta3 + k[1]*theta5 + k[2]*theta7 + k[3]*theta9;
303
304 double inv_r = r > 1e-8 ? 1.0/r : 1;
305 double cdist = r > 1e-8 ? theta_d * inv_r : 1;
306
307 Vec2d xd1 = x * cdist;
308 Vec2d xd3(xd1[0] + alpha*xd1[1], xd1[1]);
309 Vec2d final_point(xd3[0] * f[0] + c[0], xd3[1] * f[1] + c[1]);
310
311 if (undistorted.depth() == CV_32F)
312 xpf[i] = final_point;
313 else
314 xpd[i] = final_point;
315 }
316 }
317
318 //////////////////////////////////////////////////////////////////////////////////////////////////////////////
319 /// cv::fisheye::undistortPoints
320
undistortPoints(InputArray distorted,OutputArray undistorted,InputArray K,InputArray D,InputArray R,InputArray P)321 void cv::fisheye::undistortPoints( InputArray distorted, OutputArray undistorted, InputArray K, InputArray D, InputArray R, InputArray P)
322 {
323 CV_INSTRUMENT_REGION();
324
325 // will support only 2-channel data now for points
326 CV_Assert(distorted.type() == CV_32FC2 || distorted.type() == CV_64FC2);
327 undistorted.create(distorted.size(), distorted.type());
328
329 CV_Assert(P.empty() || P.size() == Size(3, 3) || P.size() == Size(4, 3));
330 CV_Assert(R.empty() || R.size() == Size(3, 3) || R.total() * R.channels() == 3);
331 CV_Assert(D.total() == 4 && K.size() == Size(3, 3) && (K.depth() == CV_32F || K.depth() == CV_64F));
332
333 cv::Vec2d f, c;
334 if (K.depth() == CV_32F)
335 {
336 Matx33f camMat = K.getMat();
337 f = Vec2f(camMat(0, 0), camMat(1, 1));
338 c = Vec2f(camMat(0, 2), camMat(1, 2));
339 }
340 else
341 {
342 Matx33d camMat = K.getMat();
343 f = Vec2d(camMat(0, 0), camMat(1, 1));
344 c = Vec2d(camMat(0, 2), camMat(1, 2));
345 }
346
347 Vec4d k = D.depth() == CV_32F ? (Vec4d)*D.getMat().ptr<Vec4f>(): *D.getMat().ptr<Vec4d>();
348
349 cv::Matx33d RR = cv::Matx33d::eye();
350 if (!R.empty() && R.total() * R.channels() == 3)
351 {
352 cv::Vec3d rvec;
353 R.getMat().convertTo(rvec, CV_64F);
354 RR = cv::Affine3d(rvec).rotation();
355 }
356 else if (!R.empty() && R.size() == Size(3, 3))
357 R.getMat().convertTo(RR, CV_64F);
358
359 if(!P.empty())
360 {
361 cv::Matx33d PP;
362 P.getMat().colRange(0, 3).convertTo(PP, CV_64F);
363 RR = PP * RR;
364 }
365
366 // start undistorting
367 const cv::Vec2f* srcf = distorted.getMat().ptr<cv::Vec2f>();
368 const cv::Vec2d* srcd = distorted.getMat().ptr<cv::Vec2d>();
369 cv::Vec2f* dstf = undistorted.getMat().ptr<cv::Vec2f>();
370 cv::Vec2d* dstd = undistorted.getMat().ptr<cv::Vec2d>();
371
372 size_t n = distorted.total();
373 int sdepth = distorted.depth();
374
375 for(size_t i = 0; i < n; i++ )
376 {
377 Vec2d pi = sdepth == CV_32F ? (Vec2d)srcf[i] : srcd[i]; // image point
378 Vec2d pw((pi[0] - c[0])/f[0], (pi[1] - c[1])/f[1]); // world point
379
380 double theta_d = sqrt(pw[0]*pw[0] + pw[1]*pw[1]);
381
382 // the current camera model is only valid up to 180 FOV
383 // for larger FOV the loop below does not converge
384 // clip values so we still get plausible results for super fisheye images > 180 grad
385 theta_d = min(max(-CV_PI/2., theta_d), CV_PI/2.);
386
387 bool converged = false;
388 double theta = theta_d;
389
390 double scale = 0.0;
391
392 if (fabs(theta_d) > 1e-8)
393 {
394 // compensate distortion iteratively
395
396 const double EPS = 1e-8; // or std::numeric_limits<double>::epsilon();
397
398 for (int j = 0; j < 10; j++)
399 {
400 double theta2 = theta*theta, theta4 = theta2*theta2, theta6 = theta4*theta2, theta8 = theta6*theta2;
401 double k0_theta2 = k[0] * theta2, k1_theta4 = k[1] * theta4, k2_theta6 = k[2] * theta6, k3_theta8 = k[3] * theta8;
402 /* new_theta = theta - theta_fix, theta_fix = f0(theta) / f0'(theta) */
403 double theta_fix = (theta * (1 + k0_theta2 + k1_theta4 + k2_theta6 + k3_theta8) - theta_d) /
404 (1 + 3*k0_theta2 + 5*k1_theta4 + 7*k2_theta6 + 9*k3_theta8);
405 theta = theta - theta_fix;
406 if (fabs(theta_fix) < EPS)
407 {
408 converged = true;
409 break;
410 }
411 }
412
413 scale = std::tan(theta) / theta_d;
414 }
415 else
416 {
417 converged = true;
418 }
419
420 // theta is monotonously increasing or decreasing depending on the sign of theta
421 // if theta has flipped, it might converge due to symmetry but on the opposite of the camera center
422 // so we can check whether theta has changed the sign during the optimization
423 bool theta_flipped = ((theta_d < 0 && theta > 0) || (theta_d > 0 && theta < 0));
424
425 if (converged && !theta_flipped)
426 {
427 Vec2d pu = pw * scale; //undistorted point
428
429 // reproject
430 Vec3d pr = RR * Vec3d(pu[0], pu[1], 1.0); // rotated point optionally multiplied by new camera matrix
431 Vec2d fi(pr[0]/pr[2], pr[1]/pr[2]); // final
432
433 if( sdepth == CV_32F )
434 dstf[i] = fi;
435 else
436 dstd[i] = fi;
437 }
438 else
439 {
440 // Vec2d fi(std::numeric_limits<double>::quiet_NaN(), std::numeric_limits<double>::quiet_NaN());
441 Vec2d fi(-1000000.0, -1000000.0);
442
443 if( sdepth == CV_32F )
444 dstf[i] = fi;
445 else
446 dstd[i] = fi;
447 }
448 }
449 }
450
451 //////////////////////////////////////////////////////////////////////////////////////////////////////////////
452 /// cv::fisheye::undistortPoints
453
initUndistortRectifyMap(InputArray K,InputArray D,InputArray R,InputArray P,const cv::Size & size,int m1type,OutputArray map1,OutputArray map2)454 void cv::fisheye::initUndistortRectifyMap( InputArray K, InputArray D, InputArray R, InputArray P,
455 const cv::Size& size, int m1type, OutputArray map1, OutputArray map2 )
456 {
457 CV_INSTRUMENT_REGION();
458
459 CV_Assert( m1type == CV_16SC2 || m1type == CV_32F || m1type <=0 );
460 map1.create( size, m1type <= 0 ? CV_16SC2 : m1type );
461 map2.create( size, map1.type() == CV_16SC2 ? CV_16UC1 : CV_32F );
462
463 CV_Assert((K.depth() == CV_32F || K.depth() == CV_64F) && (D.depth() == CV_32F || D.depth() == CV_64F));
464 CV_Assert((P.empty() || P.depth() == CV_32F || P.depth() == CV_64F) && (R.empty() || R.depth() == CV_32F || R.depth() == CV_64F));
465 CV_Assert(K.size() == Size(3, 3) && (D.empty() || D.total() == 4));
466 CV_Assert(R.empty() || R.size() == Size(3, 3) || R.total() * R.channels() == 3);
467 CV_Assert(P.empty() || P.size() == Size(3, 3) || P.size() == Size(4, 3));
468
469 cv::Vec2d f, c;
470 if (K.depth() == CV_32F)
471 {
472 Matx33f camMat = K.getMat();
473 f = Vec2f(camMat(0, 0), camMat(1, 1));
474 c = Vec2f(camMat(0, 2), camMat(1, 2));
475 }
476 else
477 {
478 Matx33d camMat = K.getMat();
479 f = Vec2d(camMat(0, 0), camMat(1, 1));
480 c = Vec2d(camMat(0, 2), camMat(1, 2));
481 }
482
483 Vec4d k = Vec4d::all(0);
484 if (!D.empty())
485 k = D.depth() == CV_32F ? (Vec4d)*D.getMat().ptr<Vec4f>(): *D.getMat().ptr<Vec4d>();
486
487 cv::Matx33d RR = cv::Matx33d::eye();
488 if (!R.empty() && R.total() * R.channels() == 3)
489 {
490 cv::Vec3d rvec;
491 R.getMat().convertTo(rvec, CV_64F);
492 RR = Affine3d(rvec).rotation();
493 }
494 else if (!R.empty() && R.size() == Size(3, 3))
495 R.getMat().convertTo(RR, CV_64F);
496
497 cv::Matx33d PP = cv::Matx33d::eye();
498 if (!P.empty())
499 P.getMat().colRange(0, 3).convertTo(PP, CV_64F);
500
501 cv::Matx33d iR = (PP * RR).inv(cv::DECOMP_SVD);
502
503 for( int i = 0; i < size.height; ++i)
504 {
505 float* m1f = map1.getMat().ptr<float>(i);
506 float* m2f = map2.getMat().ptr<float>(i);
507 short* m1 = (short*)m1f;
508 ushort* m2 = (ushort*)m2f;
509
510 double _x = i*iR(0, 1) + iR(0, 2),
511 _y = i*iR(1, 1) + iR(1, 2),
512 _w = i*iR(2, 1) + iR(2, 2);
513
514 for( int j = 0; j < size.width; ++j)
515 {
516 double u, v;
517 if( _w <= 0)
518 {
519 u = (_x > 0) ? -std::numeric_limits<double>::infinity() : std::numeric_limits<double>::infinity();
520 v = (_y > 0) ? -std::numeric_limits<double>::infinity() : std::numeric_limits<double>::infinity();
521 }
522 else
523 {
524 double x = _x/_w, y = _y/_w;
525
526 double r = sqrt(x*x + y*y);
527 double theta = atan(r);
528
529 double theta2 = theta*theta, theta4 = theta2*theta2, theta6 = theta4*theta2, theta8 = theta4*theta4;
530 double theta_d = theta * (1 + k[0]*theta2 + k[1]*theta4 + k[2]*theta6 + k[3]*theta8);
531
532 double scale = (r == 0) ? 1.0 : theta_d / r;
533 u = f[0]*x*scale + c[0];
534 v = f[1]*y*scale + c[1];
535 }
536
537 if( m1type == CV_16SC2 )
538 {
539 int iu = cv::saturate_cast<int>(u*cv::INTER_TAB_SIZE);
540 int iv = cv::saturate_cast<int>(v*cv::INTER_TAB_SIZE);
541 m1[j*2+0] = (short)(iu >> cv::INTER_BITS);
542 m1[j*2+1] = (short)(iv >> cv::INTER_BITS);
543 m2[j] = (ushort)((iv & (cv::INTER_TAB_SIZE-1))*cv::INTER_TAB_SIZE + (iu & (cv::INTER_TAB_SIZE-1)));
544 }
545 else if( m1type == CV_32FC1 )
546 {
547 m1f[j] = (float)u;
548 m2f[j] = (float)v;
549 }
550
551 _x += iR(0, 0);
552 _y += iR(1, 0);
553 _w += iR(2, 0);
554 }
555 }
556 }
557
558 //////////////////////////////////////////////////////////////////////////////////////////////////////////////
559 /// cv::fisheye::undistortImage
560
undistortImage(InputArray distorted,OutputArray undistorted,InputArray K,InputArray D,InputArray Knew,const Size & new_size)561 void cv::fisheye::undistortImage(InputArray distorted, OutputArray undistorted,
562 InputArray K, InputArray D, InputArray Knew, const Size& new_size)
563 {
564 CV_INSTRUMENT_REGION();
565
566 Size size = !new_size.empty() ? new_size : distorted.size();
567
568 cv::Mat map1, map2;
569 fisheye::initUndistortRectifyMap(K, D, cv::Matx33d::eye(), Knew, size, CV_16SC2, map1, map2 );
570 cv::remap(distorted, undistorted, map1, map2, INTER_LINEAR, BORDER_CONSTANT);
571 }
572
573
574 //////////////////////////////////////////////////////////////////////////////////////////////////////////////
575 /// cv::fisheye::estimateNewCameraMatrixForUndistortRectify
576
estimateNewCameraMatrixForUndistortRectify(InputArray K,InputArray D,const Size & image_size,InputArray R,OutputArray P,double balance,const Size & new_size,double fov_scale)577 void cv::fisheye::estimateNewCameraMatrixForUndistortRectify(InputArray K, InputArray D, const Size &image_size, InputArray R,
578 OutputArray P, double balance, const Size& new_size, double fov_scale)
579 {
580 CV_INSTRUMENT_REGION();
581
582 CV_Assert( K.size() == Size(3, 3) && (K.depth() == CV_32F || K.depth() == CV_64F));
583 CV_Assert(D.empty() || ((D.total() == 4) && (D.depth() == CV_32F || D.depth() == CV_64F)));
584
585 int w = image_size.width, h = image_size.height;
586 balance = std::min(std::max(balance, 0.0), 1.0);
587
588 cv::Mat points(1, 4, CV_64FC2);
589 Vec2d* pptr = points.ptr<Vec2d>();
590 pptr[0] = Vec2d(w/2, 0);
591 pptr[1] = Vec2d(w, h/2);
592 pptr[2] = Vec2d(w/2, h);
593 pptr[3] = Vec2d(0, h/2);
594
595 fisheye::undistortPoints(points, points, K, D, R);
596 cv::Scalar center_mass = mean(points);
597 cv::Vec2d cn(center_mass.val);
598
599 double aspect_ratio = (K.depth() == CV_32F) ? K.getMat().at<float >(0,0)/K.getMat().at<float> (1,1)
600 : K.getMat().at<double>(0,0)/K.getMat().at<double>(1,1);
601
602 // convert to identity ratio
603 cn[0] *= aspect_ratio;
604 for(size_t i = 0; i < points.total(); ++i)
605 pptr[i][1] *= aspect_ratio;
606
607 double minx = DBL_MAX, miny = DBL_MAX, maxx = -DBL_MAX, maxy = -DBL_MAX;
608 for(size_t i = 0; i < points.total(); ++i)
609 {
610 miny = std::min(miny, pptr[i][1]);
611 maxy = std::max(maxy, pptr[i][1]);
612 minx = std::min(minx, pptr[i][0]);
613 maxx = std::max(maxx, pptr[i][0]);
614 }
615
616 double f1 = w * 0.5/(cn[0] - minx);
617 double f2 = w * 0.5/(maxx - cn[0]);
618 double f3 = h * 0.5 * aspect_ratio/(cn[1] - miny);
619 double f4 = h * 0.5 * aspect_ratio/(maxy - cn[1]);
620
621 double fmin = std::min(f1, std::min(f2, std::min(f3, f4)));
622 double fmax = std::max(f1, std::max(f2, std::max(f3, f4)));
623
624 double f = balance * fmin + (1.0 - balance) * fmax;
625 f *= fov_scale > 0 ? 1.0/fov_scale : 1.0;
626
627 cv::Vec2d new_f(f, f), new_c = -cn * f + Vec2d(w, h * aspect_ratio) * 0.5;
628
629 // restore aspect ratio
630 new_f[1] /= aspect_ratio;
631 new_c[1] /= aspect_ratio;
632
633 if (!new_size.empty())
634 {
635 double rx = new_size.width /(double)image_size.width;
636 double ry = new_size.height/(double)image_size.height;
637
638 new_f[0] *= rx; new_f[1] *= ry;
639 new_c[0] *= rx; new_c[1] *= ry;
640 }
641
642 Mat(Matx33d(new_f[0], 0, new_c[0],
643 0, new_f[1], new_c[1],
644 0, 0, 1)).convertTo(P, P.empty() ? K.type() : P.type());
645 }
646
647
648 //////////////////////////////////////////////////////////////////////////////////////////////////////////////
649 /// cv::fisheye::stereoRectify
650
stereoRectify(InputArray K1,InputArray D1,InputArray K2,InputArray D2,const Size & imageSize,InputArray _R,InputArray _tvec,OutputArray R1,OutputArray R2,OutputArray P1,OutputArray P2,OutputArray Q,int flags,const Size & newImageSize,double balance,double fov_scale)651 void cv::fisheye::stereoRectify( InputArray K1, InputArray D1, InputArray K2, InputArray D2, const Size& imageSize,
652 InputArray _R, InputArray _tvec, OutputArray R1, OutputArray R2, OutputArray P1, OutputArray P2,
653 OutputArray Q, int flags, const Size& newImageSize, double balance, double fov_scale)
654 {
655 CV_INSTRUMENT_REGION();
656
657 CV_Assert((_R.size() == Size(3, 3) || _R.total() * _R.channels() == 3) && (_R.depth() == CV_32F || _R.depth() == CV_64F));
658 CV_Assert(_tvec.total() * _tvec.channels() == 3 && (_tvec.depth() == CV_32F || _tvec.depth() == CV_64F));
659
660
661 cv::Mat aaa = _tvec.getMat().reshape(3, 1);
662
663 Vec3d rvec; // Rodrigues vector
664 if (_R.size() == Size(3, 3))
665 {
666 cv::Matx33d rmat;
667 _R.getMat().convertTo(rmat, CV_64F);
668 rvec = Affine3d(rmat).rvec();
669 }
670 else if (_R.total() * _R.channels() == 3)
671 _R.getMat().convertTo(rvec, CV_64F);
672
673 Vec3d tvec;
674 _tvec.getMat().convertTo(tvec, CV_64F);
675
676 // rectification algorithm
677 rvec *= -0.5; // get average rotation
678
679 Matx33d r_r;
680 Rodrigues(rvec, r_r); // rotate cameras to same orientation by averaging
681
682 Vec3d t = r_r * tvec;
683 Vec3d uu(t[0] > 0 ? 1 : -1, 0, 0);
684
685 // calculate global Z rotation
686 Vec3d ww = t.cross(uu);
687 double nw = norm(ww);
688 if (nw > 0.0)
689 ww *= acos(fabs(t[0])/cv::norm(t))/nw;
690
691 Matx33d wr;
692 Rodrigues(ww, wr);
693
694 // apply to both views
695 Matx33d ri1 = wr * r_r.t();
696 Mat(ri1, false).convertTo(R1, R1.empty() ? CV_64F : R1.type());
697 Matx33d ri2 = wr * r_r;
698 Mat(ri2, false).convertTo(R2, R2.empty() ? CV_64F : R2.type());
699 Vec3d tnew = ri2 * tvec;
700
701 // calculate projection/camera matrices. these contain the relevant rectified image internal params (fx, fy=fx, cx, cy)
702 Matx33d newK1, newK2;
703 estimateNewCameraMatrixForUndistortRectify(K1, D1, imageSize, R1, newK1, balance, newImageSize, fov_scale);
704 estimateNewCameraMatrixForUndistortRectify(K2, D2, imageSize, R2, newK2, balance, newImageSize, fov_scale);
705
706 double fc_new = std::min(newK1(1,1), newK2(1,1));
707 Point2d cc_new[2] = { Vec2d(newK1(0, 2), newK1(1, 2)), Vec2d(newK2(0, 2), newK2(1, 2)) };
708
709 // Vertical focal length must be the same for both images to keep the epipolar constraint use fy for fx also.
710 // For simplicity, set the principal points for both cameras to be the average
711 // of the two principal points (either one of or both x- and y- coordinates)
712 if( flags & cv::CALIB_ZERO_DISPARITY )
713 cc_new[0] = cc_new[1] = (cc_new[0] + cc_new[1]) * 0.5;
714 else
715 cc_new[0].y = cc_new[1].y = (cc_new[0].y + cc_new[1].y)*0.5;
716
717 Mat(Matx34d(fc_new, 0, cc_new[0].x, 0,
718 0, fc_new, cc_new[0].y, 0,
719 0, 0, 1, 0), false).convertTo(P1, P1.empty() ? CV_64F : P1.type());
720
721 Mat(Matx34d(fc_new, 0, cc_new[1].x, tnew[0]*fc_new, // baseline * focal length;,
722 0, fc_new, cc_new[1].y, 0,
723 0, 0, 1, 0), false).convertTo(P2, P2.empty() ? CV_64F : P2.type());
724
725 if (Q.needed())
726 Mat(Matx44d(1, 0, 0, -cc_new[0].x,
727 0, 1, 0, -cc_new[0].y,
728 0, 0, 0, fc_new,
729 0, 0, -1./tnew[0], (cc_new[0].x - cc_new[1].x)/tnew[0]), false).convertTo(Q, Q.empty() ? CV_64F : Q.depth());
730 }
731
732 //////////////////////////////////////////////////////////////////////////////////////////////////////////////
733 /// cv::fisheye::calibrate
734
calibrate(InputArrayOfArrays objectPoints,InputArrayOfArrays imagePoints,const Size & image_size,InputOutputArray K,InputOutputArray D,OutputArrayOfArrays rvecs,OutputArrayOfArrays tvecs,int flags,cv::TermCriteria criteria)735 double cv::fisheye::calibrate(InputArrayOfArrays objectPoints, InputArrayOfArrays imagePoints, const Size& image_size,
736 InputOutputArray K, InputOutputArray D, OutputArrayOfArrays rvecs, OutputArrayOfArrays tvecs,
737 int flags , cv::TermCriteria criteria)
738 {
739 CV_INSTRUMENT_REGION();
740
741 CV_Assert(!objectPoints.empty() && !imagePoints.empty() && objectPoints.total() == imagePoints.total());
742 CV_Assert(objectPoints.type() == CV_32FC3 || objectPoints.type() == CV_64FC3);
743 CV_Assert(imagePoints.type() == CV_32FC2 || imagePoints.type() == CV_64FC2);
744 CV_Assert(K.empty() || (K.size() == Size(3,3)));
745 CV_Assert(D.empty() || (D.total() == 4));
746 CV_Assert(rvecs.empty() || (rvecs.channels() == 3));
747 CV_Assert(tvecs.empty() || (tvecs.channels() == 3));
748
749 CV_Assert((!K.empty() && !D.empty()) || !(flags & CALIB_USE_INTRINSIC_GUESS));
750
751 using namespace cv::internal;
752 //-------------------------------Initialization
753 IntrinsicParams finalParam;
754 IntrinsicParams currentParam;
755 IntrinsicParams errors;
756
757 finalParam.isEstimate[0] = flags & CALIB_FIX_FOCAL_LENGTH ? 0 : 1;
758 finalParam.isEstimate[1] = flags & CALIB_FIX_FOCAL_LENGTH ? 0 : 1;
759 finalParam.isEstimate[2] = flags & CALIB_FIX_PRINCIPAL_POINT ? 0 : 1;
760 finalParam.isEstimate[3] = flags & CALIB_FIX_PRINCIPAL_POINT ? 0 : 1;
761 finalParam.isEstimate[4] = flags & CALIB_FIX_SKEW ? 0 : 1;
762 finalParam.isEstimate[5] = flags & CALIB_FIX_K1 ? 0 : 1;
763 finalParam.isEstimate[6] = flags & CALIB_FIX_K2 ? 0 : 1;
764 finalParam.isEstimate[7] = flags & CALIB_FIX_K3 ? 0 : 1;
765 finalParam.isEstimate[8] = flags & CALIB_FIX_K4 ? 0 : 1;
766
767 const int recompute_extrinsic = flags & CALIB_RECOMPUTE_EXTRINSIC ? 1: 0;
768 const int check_cond = flags & CALIB_CHECK_COND ? 1 : 0;
769
770 const double alpha_smooth = 0.4;
771 const double thresh_cond = 1e6;
772 double change = 1;
773 Vec2d err_std;
774
775 Matx33d _K;
776 Vec4d _D;
777 if (flags & CALIB_USE_INTRINSIC_GUESS)
778 {
779 K.getMat().convertTo(_K, CV_64FC1);
780 D.getMat().convertTo(_D, CV_64FC1);
781 finalParam.Init(Vec2d(_K(0,0), _K(1, 1)),
782 Vec2d(_K(0,2), _K(1, 2)),
783 Vec4d(flags & CALIB_FIX_K1 ? 0 : _D[0],
784 flags & CALIB_FIX_K2 ? 0 : _D[1],
785 flags & CALIB_FIX_K3 ? 0 : _D[2],
786 flags & CALIB_FIX_K4 ? 0 : _D[3]),
787 _K(0, 1) / _K(0, 0));
788 }
789 else
790 {
791 finalParam.Init(Vec2d(max(image_size.width, image_size.height) / CV_PI, max(image_size.width, image_size.height) / CV_PI),
792 Vec2d(image_size.width / 2.0 - 0.5, image_size.height / 2.0 - 0.5));
793 }
794
795 errors.isEstimate = finalParam.isEstimate;
796
797 std::vector<Vec3d> omc(objectPoints.total()), Tc(objectPoints.total());
798
799 CalibrateExtrinsics(objectPoints, imagePoints, finalParam, check_cond, thresh_cond, omc, Tc);
800
801
802 //-------------------------------Optimization
803 for(int iter = 0; iter < std::numeric_limits<int>::max(); ++iter)
804 {
805 if ((criteria.type == 1 && iter >= criteria.maxCount) ||
806 (criteria.type == 2 && change <= criteria.epsilon) ||
807 (criteria.type == 3 && (change <= criteria.epsilon || iter >= criteria.maxCount)))
808 break;
809
810 double alpha_smooth2 = 1 - std::pow(1 - alpha_smooth, iter + 1.0);
811
812 Mat JJ2, ex3;
813 ComputeJacobians(objectPoints, imagePoints, finalParam, omc, Tc, check_cond,thresh_cond, JJ2, ex3);
814
815 Mat G;
816 solve(JJ2, ex3, G);
817 currentParam = finalParam + alpha_smooth2*G;
818
819 change = norm(Vec4d(currentParam.f[0], currentParam.f[1], currentParam.c[0], currentParam.c[1]) -
820 Vec4d(finalParam.f[0], finalParam.f[1], finalParam.c[0], finalParam.c[1]))
821 / norm(Vec4d(currentParam.f[0], currentParam.f[1], currentParam.c[0], currentParam.c[1]));
822
823 finalParam = currentParam;
824
825 if (recompute_extrinsic)
826 {
827 CalibrateExtrinsics(objectPoints, imagePoints, finalParam, check_cond,
828 thresh_cond, omc, Tc);
829 }
830 }
831
832 //-------------------------------Validation
833 double rms;
834 EstimateUncertainties(objectPoints, imagePoints, finalParam, omc, Tc, errors, err_std, thresh_cond,
835 check_cond, rms);
836
837 //-------------------------------
838 _K = Matx33d(finalParam.f[0], finalParam.f[0] * finalParam.alpha, finalParam.c[0],
839 0, finalParam.f[1], finalParam.c[1],
840 0, 0, 1);
841
842 if (K.needed()) cv::Mat(_K).convertTo(K, K.empty() ? CV_64FC1 : K.type());
843 if (D.needed()) cv::Mat(finalParam.k).convertTo(D, D.empty() ? CV_64FC1 : D.type());
844 if (rvecs.isMatVector())
845 {
846 int N = (int)objectPoints.total();
847
848 if(rvecs.empty())
849 rvecs.create(N, 1, CV_64FC3);
850
851 if(tvecs.empty())
852 tvecs.create(N, 1, CV_64FC3);
853
854 for(int i = 0; i < N; i++ )
855 {
856 rvecs.create(3, 1, CV_64F, i, true);
857 tvecs.create(3, 1, CV_64F, i, true);
858 memcpy(rvecs.getMat(i).ptr(), omc[i].val, sizeof(Vec3d));
859 memcpy(tvecs.getMat(i).ptr(), Tc[i].val, sizeof(Vec3d));
860 }
861 }
862 else
863 {
864 if (rvecs.needed()) cv::Mat(omc).convertTo(rvecs, rvecs.empty() ? CV_64FC3 : rvecs.type());
865 if (tvecs.needed()) cv::Mat(Tc).convertTo(tvecs, tvecs.empty() ? CV_64FC3 : tvecs.type());
866 }
867
868 return rms;
869 }
870
871 //////////////////////////////////////////////////////////////////////////////////////////////////////////////
872 /// cv::fisheye::stereoCalibrate
873
stereoCalibrate(InputArrayOfArrays objectPoints,InputArrayOfArrays imagePoints1,InputArrayOfArrays imagePoints2,InputOutputArray K1,InputOutputArray D1,InputOutputArray K2,InputOutputArray D2,Size imageSize,OutputArray R,OutputArray T,int flags,TermCriteria criteria)874 double cv::fisheye::stereoCalibrate(InputArrayOfArrays objectPoints, InputArrayOfArrays imagePoints1, InputArrayOfArrays imagePoints2,
875 InputOutputArray K1, InputOutputArray D1, InputOutputArray K2, InputOutputArray D2, Size imageSize,
876 OutputArray R, OutputArray T, int flags, TermCriteria criteria)
877 {
878 CV_INSTRUMENT_REGION();
879
880 CV_Assert(!objectPoints.empty() && !imagePoints1.empty() && !imagePoints2.empty());
881 CV_Assert(objectPoints.total() == imagePoints1.total() || imagePoints1.total() == imagePoints2.total());
882 CV_Assert(objectPoints.type() == CV_32FC3 || objectPoints.type() == CV_64FC3);
883 CV_Assert(imagePoints1.type() == CV_32FC2 || imagePoints1.type() == CV_64FC2);
884 CV_Assert(imagePoints2.type() == CV_32FC2 || imagePoints2.type() == CV_64FC2);
885
886 CV_Assert(K1.empty() || (K1.size() == Size(3,3)));
887 CV_Assert(D1.empty() || (D1.total() == 4));
888 CV_Assert(K2.empty() || (K2.size() == Size(3,3)));
889 CV_Assert(D2.empty() || (D2.total() == 4));
890
891 CV_Assert((!K1.empty() && !K2.empty() && !D1.empty() && !D2.empty()) || !(flags & CALIB_FIX_INTRINSIC));
892
893 //-------------------------------Initialization
894
895 const int threshold = 50;
896 const double thresh_cond = 1e6;
897 const int check_cond = 1;
898
899 int n_points = (int)objectPoints.getMat(0).total();
900 int n_images = (int)objectPoints.total();
901
902 double change = 1;
903
904 cv::internal::IntrinsicParams intrinsicLeft;
905 cv::internal::IntrinsicParams intrinsicRight;
906
907 cv::internal::IntrinsicParams intrinsicLeft_errors;
908 cv::internal::IntrinsicParams intrinsicRight_errors;
909
910 Matx33d _K1, _K2;
911 Vec4d _D1, _D2;
912 if (!K1.empty()) K1.getMat().convertTo(_K1, CV_64FC1);
913 if (!D1.empty()) D1.getMat().convertTo(_D1, CV_64FC1);
914 if (!K2.empty()) K2.getMat().convertTo(_K2, CV_64FC1);
915 if (!D2.empty()) D2.getMat().convertTo(_D2, CV_64FC1);
916
917 std::vector<Vec3d> rvecs1(n_images), tvecs1(n_images), rvecs2(n_images), tvecs2(n_images);
918
919 if (!(flags & CALIB_FIX_INTRINSIC))
920 {
921 calibrate(objectPoints, imagePoints1, imageSize, _K1, _D1, rvecs1, tvecs1, flags, TermCriteria(3, 20, 1e-6));
922 calibrate(objectPoints, imagePoints2, imageSize, _K2, _D2, rvecs2, tvecs2, flags, TermCriteria(3, 20, 1e-6));
923 }
924
925 intrinsicLeft.Init(Vec2d(_K1(0,0), _K1(1, 1)), Vec2d(_K1(0,2), _K1(1, 2)),
926 Vec4d(_D1[0], _D1[1], _D1[2], _D1[3]), _K1(0, 1) / _K1(0, 0));
927
928 intrinsicRight.Init(Vec2d(_K2(0,0), _K2(1, 1)), Vec2d(_K2(0,2), _K2(1, 2)),
929 Vec4d(_D2[0], _D2[1], _D2[2], _D2[3]), _K2(0, 1) / _K2(0, 0));
930
931 if ((flags & CALIB_FIX_INTRINSIC))
932 {
933 cv::internal::CalibrateExtrinsics(objectPoints, imagePoints1, intrinsicLeft, check_cond, thresh_cond, rvecs1, tvecs1);
934 cv::internal::CalibrateExtrinsics(objectPoints, imagePoints2, intrinsicRight, check_cond, thresh_cond, rvecs2, tvecs2);
935 }
936
937 intrinsicLeft.isEstimate[0] = flags & CALIB_FIX_INTRINSIC ? 0 : 1;
938 intrinsicLeft.isEstimate[1] = flags & CALIB_FIX_INTRINSIC ? 0 : 1;
939 intrinsicLeft.isEstimate[2] = flags & CALIB_FIX_INTRINSIC ? 0 : 1;
940 intrinsicLeft.isEstimate[3] = flags & CALIB_FIX_INTRINSIC ? 0 : 1;
941 intrinsicLeft.isEstimate[4] = flags & (CALIB_FIX_SKEW | CALIB_FIX_INTRINSIC) ? 0 : 1;
942 intrinsicLeft.isEstimate[5] = flags & (CALIB_FIX_K1 | CALIB_FIX_INTRINSIC) ? 0 : 1;
943 intrinsicLeft.isEstimate[6] = flags & (CALIB_FIX_K2 | CALIB_FIX_INTRINSIC) ? 0 : 1;
944 intrinsicLeft.isEstimate[7] = flags & (CALIB_FIX_K3 | CALIB_FIX_INTRINSIC) ? 0 : 1;
945 intrinsicLeft.isEstimate[8] = flags & (CALIB_FIX_K4 | CALIB_FIX_INTRINSIC) ? 0 : 1;
946
947 intrinsicRight.isEstimate[0] = flags & CALIB_FIX_INTRINSIC ? 0 : 1;
948 intrinsicRight.isEstimate[1] = flags & CALIB_FIX_INTRINSIC ? 0 : 1;
949 intrinsicRight.isEstimate[2] = flags & CALIB_FIX_INTRINSIC ? 0 : 1;
950 intrinsicRight.isEstimate[3] = flags & CALIB_FIX_INTRINSIC ? 0 : 1;
951 intrinsicRight.isEstimate[4] = flags & (CALIB_FIX_SKEW | CALIB_FIX_INTRINSIC) ? 0 : 1;
952 intrinsicRight.isEstimate[5] = flags & (CALIB_FIX_K1 | CALIB_FIX_INTRINSIC) ? 0 : 1;
953 intrinsicRight.isEstimate[6] = flags & (CALIB_FIX_K2 | CALIB_FIX_INTRINSIC) ? 0 : 1;
954 intrinsicRight.isEstimate[7] = flags & (CALIB_FIX_K3 | CALIB_FIX_INTRINSIC) ? 0 : 1;
955 intrinsicRight.isEstimate[8] = flags & (CALIB_FIX_K4 | CALIB_FIX_INTRINSIC) ? 0 : 1;
956
957 intrinsicLeft_errors.isEstimate = intrinsicLeft.isEstimate;
958 intrinsicRight_errors.isEstimate = intrinsicRight.isEstimate;
959
960 std::vector<uchar> selectedParams;
961 std::vector<uchar> tmp(6 * (n_images + 1), 1);
962 selectedParams.insert(selectedParams.end(), intrinsicLeft.isEstimate.begin(), intrinsicLeft.isEstimate.end());
963 selectedParams.insert(selectedParams.end(), intrinsicRight.isEstimate.begin(), intrinsicRight.isEstimate.end());
964 selectedParams.insert(selectedParams.end(), tmp.begin(), tmp.end());
965
966 //Init values for rotation and translation between two views
967 cv::Mat om_list(1, n_images, CV_64FC3), T_list(1, n_images, CV_64FC3);
968 cv::Mat om_ref, R_ref, T_ref, R1, R2;
969 for (int image_idx = 0; image_idx < n_images; ++image_idx)
970 {
971 cv::Rodrigues(rvecs1[image_idx], R1);
972 cv::Rodrigues(rvecs2[image_idx], R2);
973 R_ref = R2 * R1.t();
974 T_ref = cv::Mat(tvecs2[image_idx]) - R_ref * cv::Mat(tvecs1[image_idx]);
975 cv::Rodrigues(R_ref, om_ref);
976 om_ref.reshape(3, 1).copyTo(om_list.col(image_idx));
977 T_ref.reshape(3, 1).copyTo(T_list.col(image_idx));
978 }
979 cv::Vec3d omcur = cv::internal::median3d(om_list);
980 cv::Vec3d Tcur = cv::internal::median3d(T_list);
981
982 cv::Mat J = cv::Mat::zeros(4 * n_points * n_images, 18 + 6 * (n_images + 1), CV_64FC1),
983 e = cv::Mat::zeros(4 * n_points * n_images, 1, CV_64FC1), Jkk, ekk;
984
985 for(int iter = 0; ; ++iter)
986 {
987 if ((criteria.type == 1 && iter >= criteria.maxCount) ||
988 (criteria.type == 2 && change <= criteria.epsilon) ||
989 (criteria.type == 3 && (change <= criteria.epsilon || iter >= criteria.maxCount)))
990 break;
991
992 J.create(4 * n_points * n_images, 18 + 6 * (n_images + 1), CV_64FC1);
993 e.create(4 * n_points * n_images, 1, CV_64FC1);
994 Jkk.create(4 * n_points, 18 + 6 * (n_images + 1), CV_64FC1);
995 ekk.create(4 * n_points, 1, CV_64FC1);
996
997 cv::Mat omr, Tr, domrdomckk, domrdTckk, domrdom, domrdT, dTrdomckk, dTrdTckk, dTrdom, dTrdT;
998
999 for (int image_idx = 0; image_idx < n_images; ++image_idx)
1000 {
1001 Jkk = cv::Mat::zeros(4 * n_points, 18 + 6 * (n_images + 1), CV_64FC1);
1002
1003 cv::Mat object = objectPoints.getMat(image_idx).clone();
1004 cv::Mat imageLeft = imagePoints1.getMat(image_idx).clone();
1005 cv::Mat imageRight = imagePoints2.getMat(image_idx).clone();
1006 cv::Mat jacobians, projected;
1007
1008 //left camera jacobian
1009 cv::Mat rvec = cv::Mat(rvecs1[image_idx]);
1010 cv::Mat tvec = cv::Mat(tvecs1[image_idx]);
1011 cv::internal::projectPoints(object, projected, rvec, tvec, intrinsicLeft, jacobians);
1012 cv::Mat(cv::Mat((imageLeft - projected).t()).reshape(1, 1).t()).copyTo(ekk.rowRange(0, 2 * n_points));
1013 jacobians.colRange(8, 11).copyTo(Jkk.colRange(24 + image_idx * 6, 27 + image_idx * 6).rowRange(0, 2 * n_points));
1014 jacobians.colRange(11, 14).copyTo(Jkk.colRange(27 + image_idx * 6, 30 + image_idx * 6).rowRange(0, 2 * n_points));
1015 jacobians.colRange(0, 2).copyTo(Jkk.colRange(0, 2).rowRange(0, 2 * n_points));
1016 jacobians.colRange(2, 4).copyTo(Jkk.colRange(2, 4).rowRange(0, 2 * n_points));
1017 jacobians.colRange(4, 8).copyTo(Jkk.colRange(5, 9).rowRange(0, 2 * n_points));
1018 jacobians.col(14).copyTo(Jkk.col(4).rowRange(0, 2 * n_points));
1019
1020 //right camera jacobian
1021 cv::internal::compose_motion(rvec, tvec, omcur, Tcur, omr, Tr, domrdomckk, domrdTckk, domrdom, domrdT, dTrdomckk, dTrdTckk, dTrdom, dTrdT);
1022 rvec = cv::Mat(rvecs2[image_idx]);
1023 tvec = cv::Mat(tvecs2[image_idx]);
1024
1025 cv::internal::projectPoints(object, projected, omr, Tr, intrinsicRight, jacobians);
1026 cv::Mat(cv::Mat((imageRight - projected).t()).reshape(1, 1).t()).copyTo(ekk.rowRange(2 * n_points, 4 * n_points));
1027 cv::Mat dxrdom = jacobians.colRange(8, 11) * domrdom + jacobians.colRange(11, 14) * dTrdom;
1028 cv::Mat dxrdT = jacobians.colRange(8, 11) * domrdT + jacobians.colRange(11, 14)* dTrdT;
1029 cv::Mat dxrdomckk = jacobians.colRange(8, 11) * domrdomckk + jacobians.colRange(11, 14) * dTrdomckk;
1030 cv::Mat dxrdTckk = jacobians.colRange(8, 11) * domrdTckk + jacobians.colRange(11, 14) * dTrdTckk;
1031
1032 dxrdom.copyTo(Jkk.colRange(18, 21).rowRange(2 * n_points, 4 * n_points));
1033 dxrdT.copyTo(Jkk.colRange(21, 24).rowRange(2 * n_points, 4 * n_points));
1034 dxrdomckk.copyTo(Jkk.colRange(24 + image_idx * 6, 27 + image_idx * 6).rowRange(2 * n_points, 4 * n_points));
1035 dxrdTckk.copyTo(Jkk.colRange(27 + image_idx * 6, 30 + image_idx * 6).rowRange(2 * n_points, 4 * n_points));
1036 jacobians.colRange(0, 2).copyTo(Jkk.colRange(9 + 0, 9 + 2).rowRange(2 * n_points, 4 * n_points));
1037 jacobians.colRange(2, 4).copyTo(Jkk.colRange(9 + 2, 9 + 4).rowRange(2 * n_points, 4 * n_points));
1038 jacobians.colRange(4, 8).copyTo(Jkk.colRange(9 + 5, 9 + 9).rowRange(2 * n_points, 4 * n_points));
1039 jacobians.col(14).copyTo(Jkk.col(9 + 4).rowRange(2 * n_points, 4 * n_points));
1040
1041 //check goodness of sterepair
1042 double abs_max = 0;
1043 for (int i = 0; i < 4 * n_points; i++)
1044 {
1045 if (fabs(ekk.at<double>(i)) > abs_max)
1046 {
1047 abs_max = fabs(ekk.at<double>(i));
1048 }
1049 }
1050
1051 CV_Assert(abs_max < threshold); // bad stereo pair
1052
1053 Jkk.copyTo(J.rowRange(image_idx * 4 * n_points, (image_idx + 1) * 4 * n_points));
1054 ekk.copyTo(e.rowRange(image_idx * 4 * n_points, (image_idx + 1) * 4 * n_points));
1055 }
1056
1057 cv::Vec6d oldTom(Tcur[0], Tcur[1], Tcur[2], omcur[0], omcur[1], omcur[2]);
1058
1059 //update all parameters
1060 cv::subMatrix(J, J, selectedParams, std::vector<uchar>(J.rows, 1));
1061 int a = cv::countNonZero(intrinsicLeft.isEstimate);
1062 int b = cv::countNonZero(intrinsicRight.isEstimate);
1063 cv::Mat deltas;
1064 solve(J.t() * J, J.t()*e, deltas);
1065 if (a > 0)
1066 intrinsicLeft = intrinsicLeft + deltas.rowRange(0, a);
1067 if (b > 0)
1068 intrinsicRight = intrinsicRight + deltas.rowRange(a, a + b);
1069 omcur = omcur + cv::Vec3d(deltas.rowRange(a + b, a + b + 3));
1070 Tcur = Tcur + cv::Vec3d(deltas.rowRange(a + b + 3, a + b + 6));
1071 for (int image_idx = 0; image_idx < n_images; ++image_idx)
1072 {
1073 rvecs1[image_idx] = cv::Mat(cv::Mat(rvecs1[image_idx]) + deltas.rowRange(a + b + 6 + image_idx * 6, a + b + 9 + image_idx * 6));
1074 tvecs1[image_idx] = cv::Mat(cv::Mat(tvecs1[image_idx]) + deltas.rowRange(a + b + 9 + image_idx * 6, a + b + 12 + image_idx * 6));
1075 }
1076
1077 cv::Vec6d newTom(Tcur[0], Tcur[1], Tcur[2], omcur[0], omcur[1], omcur[2]);
1078 change = cv::norm(newTom - oldTom) / cv::norm(newTom);
1079 }
1080
1081 double rms = 0;
1082 const Vec2d* ptr_e = e.ptr<Vec2d>();
1083 for (size_t i = 0; i < e.total() / 2; i++)
1084 {
1085 rms += ptr_e[i][0] * ptr_e[i][0] + ptr_e[i][1] * ptr_e[i][1];
1086 }
1087
1088 rms /= ((double)e.total() / 2.0);
1089 rms = sqrt(rms);
1090
1091 _K1 = Matx33d(intrinsicLeft.f[0], intrinsicLeft.f[0] * intrinsicLeft.alpha, intrinsicLeft.c[0],
1092 0, intrinsicLeft.f[1], intrinsicLeft.c[1],
1093 0, 0, 1);
1094
1095 _K2 = Matx33d(intrinsicRight.f[0], intrinsicRight.f[0] * intrinsicRight.alpha, intrinsicRight.c[0],
1096 0, intrinsicRight.f[1], intrinsicRight.c[1],
1097 0, 0, 1);
1098
1099 Mat _R;
1100 Rodrigues(omcur, _R);
1101
1102 if (K1.needed()) cv::Mat(_K1).convertTo(K1, K1.empty() ? CV_64FC1 : K1.type());
1103 if (K2.needed()) cv::Mat(_K2).convertTo(K2, K2.empty() ? CV_64FC1 : K2.type());
1104 if (D1.needed()) cv::Mat(intrinsicLeft.k).convertTo(D1, D1.empty() ? CV_64FC1 : D1.type());
1105 if (D2.needed()) cv::Mat(intrinsicRight.k).convertTo(D2, D2.empty() ? CV_64FC1 : D2.type());
1106 if (R.needed()) _R.convertTo(R, R.empty() ? CV_64FC1 : R.type());
1107 if (T.needed()) cv::Mat(Tcur).convertTo(T, T.empty() ? CV_64FC1 : T.type());
1108
1109 return rms;
1110 }
1111
1112 namespace cv{ namespace {
subMatrix(const Mat & src,Mat & dst,const std::vector<uchar> & cols,const std::vector<uchar> & rows)1113 void subMatrix(const Mat& src, Mat& dst, const std::vector<uchar>& cols, const std::vector<uchar>& rows)
1114 {
1115 CV_Assert(src.channels() == 1);
1116
1117 int nonzeros_cols = cv::countNonZero(cols);
1118 Mat tmp(src.rows, nonzeros_cols, CV_64F);
1119
1120 for (int i = 0, j = 0; i < (int)cols.size(); i++)
1121 {
1122 if (cols[i])
1123 {
1124 src.col(i).copyTo(tmp.col(j++));
1125 }
1126 }
1127
1128 int nonzeros_rows = cv::countNonZero(rows);
1129 dst.create(nonzeros_rows, nonzeros_cols, CV_64F);
1130 for (int i = 0, j = 0; i < (int)rows.size(); i++)
1131 {
1132 if (rows[i])
1133 {
1134 tmp.row(i).copyTo(dst.row(j++));
1135 }
1136 }
1137 }
1138
1139 }}
1140
IntrinsicParams()1141 cv::internal::IntrinsicParams::IntrinsicParams():
1142 f(Vec2d::all(0)), c(Vec2d::all(0)), k(Vec4d::all(0)), alpha(0), isEstimate(9,0)
1143 {
1144 }
1145
IntrinsicParams(Vec2d _f,Vec2d _c,Vec4d _k,double _alpha)1146 cv::internal::IntrinsicParams::IntrinsicParams(Vec2d _f, Vec2d _c, Vec4d _k, double _alpha):
1147 f(_f), c(_c), k(_k), alpha(_alpha), isEstimate(9,0)
1148 {
1149 }
1150
operator +(const Mat & a)1151 cv::internal::IntrinsicParams cv::internal::IntrinsicParams::operator+(const Mat& a)
1152 {
1153 CV_Assert(a.type() == CV_64FC1);
1154 IntrinsicParams tmp;
1155 const double* ptr = a.ptr<double>();
1156
1157 int j = 0;
1158 tmp.f[0] = this->f[0] + (isEstimate[0] ? ptr[j++] : 0);
1159 tmp.f[1] = this->f[1] + (isEstimate[1] ? ptr[j++] : 0);
1160 tmp.c[0] = this->c[0] + (isEstimate[2] ? ptr[j++] : 0);
1161 tmp.c[1] = this->c[1] + (isEstimate[3] ? ptr[j++] : 0);
1162 tmp.alpha = this->alpha + (isEstimate[4] ? ptr[j++] : 0);
1163 tmp.k[0] = this->k[0] + (isEstimate[5] ? ptr[j++] : 0);
1164 tmp.k[1] = this->k[1] + (isEstimate[6] ? ptr[j++] : 0);
1165 tmp.k[2] = this->k[2] + (isEstimate[7] ? ptr[j++] : 0);
1166 tmp.k[3] = this->k[3] + (isEstimate[8] ? ptr[j++] : 0);
1167
1168 tmp.isEstimate = isEstimate;
1169 return tmp;
1170 }
1171
operator =(const Mat & a)1172 cv::internal::IntrinsicParams& cv::internal::IntrinsicParams::operator =(const Mat& a)
1173 {
1174 CV_Assert(a.type() == CV_64FC1);
1175 const double* ptr = a.ptr<double>();
1176
1177 int j = 0;
1178
1179 this->f[0] = isEstimate[0] ? ptr[j++] : 0;
1180 this->f[1] = isEstimate[1] ? ptr[j++] : 0;
1181 this->c[0] = isEstimate[2] ? ptr[j++] : 0;
1182 this->c[1] = isEstimate[3] ? ptr[j++] : 0;
1183 this->alpha = isEstimate[4] ? ptr[j++] : 0;
1184 this->k[0] = isEstimate[5] ? ptr[j++] : 0;
1185 this->k[1] = isEstimate[6] ? ptr[j++] : 0;
1186 this->k[2] = isEstimate[7] ? ptr[j++] : 0;
1187 this->k[3] = isEstimate[8] ? ptr[j++] : 0;
1188
1189 return *this;
1190 }
1191
Init(const cv::Vec2d & _f,const cv::Vec2d & _c,const cv::Vec4d & _k,const double & _alpha)1192 void cv::internal::IntrinsicParams::Init(const cv::Vec2d& _f, const cv::Vec2d& _c, const cv::Vec4d& _k, const double& _alpha)
1193 {
1194 this->c = _c;
1195 this->f = _f;
1196 this->k = _k;
1197 this->alpha = _alpha;
1198 }
1199
projectPoints(cv::InputArray objectPoints,cv::OutputArray imagePoints,cv::InputArray _rvec,cv::InputArray _tvec,const IntrinsicParams & param,cv::OutputArray jacobian)1200 void cv::internal::projectPoints(cv::InputArray objectPoints, cv::OutputArray imagePoints,
1201 cv::InputArray _rvec,cv::InputArray _tvec,
1202 const IntrinsicParams& param, cv::OutputArray jacobian)
1203 {
1204 CV_INSTRUMENT_REGION();
1205
1206 CV_Assert(!objectPoints.empty() && (objectPoints.type() == CV_32FC3 || objectPoints.type() == CV_64FC3));
1207 Matx33d K(param.f[0], param.f[0] * param.alpha, param.c[0],
1208 0, param.f[1], param.c[1],
1209 0, 0, 1);
1210 fisheye::projectPoints(objectPoints, imagePoints, _rvec, _tvec, K, param.k, param.alpha, jacobian);
1211 }
1212
ComputeExtrinsicRefine(const Mat & imagePoints,const Mat & objectPoints,Mat & rvec,Mat & tvec,Mat & J,const int MaxIter,const IntrinsicParams & param,const double thresh_cond)1213 void cv::internal::ComputeExtrinsicRefine(const Mat& imagePoints, const Mat& objectPoints, Mat& rvec,
1214 Mat& tvec, Mat& J, const int MaxIter,
1215 const IntrinsicParams& param, const double thresh_cond)
1216 {
1217 CV_Assert(!objectPoints.empty() && objectPoints.type() == CV_64FC3);
1218 CV_Assert(!imagePoints.empty() && imagePoints.type() == CV_64FC2);
1219 CV_Assert(rvec.total() > 2 && tvec.total() > 2);
1220 Vec6d extrinsics(rvec.at<double>(0), rvec.at<double>(1), rvec.at<double>(2),
1221 tvec.at<double>(0), tvec.at<double>(1), tvec.at<double>(2));
1222 double change = 1;
1223 int iter = 0;
1224
1225 while (change > 1e-10 && iter < MaxIter)
1226 {
1227 std::vector<Point2d> x;
1228 Mat jacobians;
1229 projectPoints(objectPoints, x, rvec, tvec, param, jacobians);
1230
1231 Mat ex = imagePoints - Mat(x).t();
1232 ex = ex.reshape(1, 2);
1233
1234 J = jacobians.colRange(8, 14).clone();
1235
1236 SVD svd(J, SVD::NO_UV);
1237 double condJJ = svd.w.at<double>(0)/svd.w.at<double>(5);
1238
1239 if (condJJ > thresh_cond)
1240 change = 0;
1241 else
1242 {
1243 Vec6d param_innov;
1244 solve(J, ex.reshape(1, (int)ex.total()), param_innov, DECOMP_SVD + DECOMP_NORMAL);
1245
1246 Vec6d param_up = extrinsics + param_innov;
1247 change = norm(param_innov)/norm(param_up);
1248 extrinsics = param_up;
1249 iter = iter + 1;
1250
1251 rvec = Mat(Vec3d(extrinsics.val));
1252 tvec = Mat(Vec3d(extrinsics.val+3));
1253 }
1254 }
1255 }
1256
ComputeHomography(Mat m,Mat M)1257 cv::Mat cv::internal::ComputeHomography(Mat m, Mat M)
1258 {
1259 CV_INSTRUMENT_REGION();
1260
1261 int Np = m.cols;
1262
1263 if (m.rows < 3)
1264 {
1265 vconcat(m, Mat::ones(1, Np, CV_64FC1), m);
1266 }
1267 if (M.rows < 3)
1268 {
1269 vconcat(M, Mat::ones(1, Np, CV_64FC1), M);
1270 }
1271
1272 divide(m, Mat::ones(3, 1, CV_64FC1) * m.row(2), m);
1273 divide(M, Mat::ones(3, 1, CV_64FC1) * M.row(2), M);
1274
1275 Mat ax = m.row(0).clone();
1276 Mat ay = m.row(1).clone();
1277
1278 double mxx = mean(ax)[0];
1279 double myy = mean(ay)[0];
1280
1281 ax = ax - mxx;
1282 ay = ay - myy;
1283
1284 double scxx = mean(abs(ax))[0];
1285 double scyy = mean(abs(ay))[0];
1286
1287 Mat Hnorm (Matx33d( 1/scxx, 0.0, -mxx/scxx,
1288 0.0, 1/scyy, -myy/scyy,
1289 0.0, 0.0, 1.0 ));
1290
1291 Mat inv_Hnorm (Matx33d( scxx, 0, mxx,
1292 0, scyy, myy,
1293 0, 0, 1 ));
1294 Mat mn = Hnorm * m;
1295
1296 Mat L = Mat::zeros(2*Np, 9, CV_64FC1);
1297
1298 for (int i = 0; i < Np; ++i)
1299 {
1300 for (int j = 0; j < 3; j++)
1301 {
1302 L.at<double>(2 * i, j) = M.at<double>(j, i);
1303 L.at<double>(2 * i + 1, j + 3) = M.at<double>(j, i);
1304 L.at<double>(2 * i, j + 6) = -mn.at<double>(0,i) * M.at<double>(j, i);
1305 L.at<double>(2 * i + 1, j + 6) = -mn.at<double>(1,i) * M.at<double>(j, i);
1306 }
1307 }
1308
1309 if (Np > 4) L = L.t() * L;
1310 SVD svd(L);
1311 Mat hh = svd.vt.row(8) / svd.vt.row(8).at<double>(8);
1312 Mat Hrem = hh.reshape(1, 3);
1313 Mat H = inv_Hnorm * Hrem;
1314
1315 if (Np > 4)
1316 {
1317 Mat hhv = H.reshape(1, 9)(Rect(0, 0, 1, 8)).clone();
1318 for (int iter = 0; iter < 10; iter++)
1319 {
1320 Mat mrep = H * M;
1321 Mat J = Mat::zeros(2 * Np, 8, CV_64FC1);
1322 Mat MMM;
1323 divide(M, Mat::ones(3, 1, CV_64FC1) * mrep(Rect(0, 2, mrep.cols, 1)), MMM);
1324 divide(mrep, Mat::ones(3, 1, CV_64FC1) * mrep(Rect(0, 2, mrep.cols, 1)), mrep);
1325 Mat m_err = m(Rect(0,0, m.cols, 2)) - mrep(Rect(0,0, mrep.cols, 2));
1326 m_err = Mat(m_err.t()).reshape(1, m_err.cols * m_err.rows);
1327 Mat MMM2, MMM3;
1328 multiply(Mat::ones(3, 1, CV_64FC1) * mrep(Rect(0, 0, mrep.cols, 1)), MMM, MMM2);
1329 multiply(Mat::ones(3, 1, CV_64FC1) * mrep(Rect(0, 1, mrep.cols, 1)), MMM, MMM3);
1330
1331 for (int i = 0; i < Np; ++i)
1332 {
1333 for (int j = 0; j < 3; ++j)
1334 {
1335 J.at<double>(2 * i, j) = -MMM.at<double>(j, i);
1336 J.at<double>(2 * i + 1, j + 3) = -MMM.at<double>(j, i);
1337 }
1338
1339 for (int j = 0; j < 2; ++j)
1340 {
1341 J.at<double>(2 * i, j + 6) = MMM2.at<double>(j, i);
1342 J.at<double>(2 * i + 1, j + 6) = MMM3.at<double>(j, i);
1343 }
1344 }
1345 divide(M, Mat::ones(3, 1, CV_64FC1) * mrep(Rect(0,2,mrep.cols,1)), MMM);
1346 Mat hh_innov = (J.t() * J).inv() * (J.t()) * m_err;
1347 Mat hhv_up = hhv - hh_innov;
1348 Mat tmp;
1349 vconcat(hhv_up, Mat::ones(1,1,CV_64FC1), tmp);
1350 Mat H_up = tmp.reshape(1,3);
1351 hhv = hhv_up;
1352 H = H_up;
1353 }
1354 }
1355 return H;
1356 }
1357
NormalizePixels(const Mat & imagePoints,const IntrinsicParams & param)1358 cv::Mat cv::internal::NormalizePixels(const Mat& imagePoints, const IntrinsicParams& param)
1359 {
1360 CV_INSTRUMENT_REGION();
1361
1362 CV_Assert(!imagePoints.empty() && imagePoints.type() == CV_64FC2);
1363
1364 Mat distorted((int)imagePoints.total(), 1, CV_64FC2), undistorted;
1365 const Vec2d* ptr = imagePoints.ptr<Vec2d>();
1366 Vec2d* ptr_d = distorted.ptr<Vec2d>();
1367 for (size_t i = 0; i < imagePoints.total(); ++i)
1368 {
1369 ptr_d[i] = (ptr[i] - param.c).mul(Vec2d(1.0 / param.f[0], 1.0 / param.f[1]));
1370 ptr_d[i][0] -= param.alpha * ptr_d[i][1];
1371 }
1372 cv::fisheye::undistortPoints(distorted, undistorted, Matx33d::eye(), param.k);
1373 return undistorted;
1374 }
1375
InitExtrinsics(const Mat & _imagePoints,const Mat & _objectPoints,const IntrinsicParams & param,Mat & omckk,Mat & Tckk)1376 void cv::internal::InitExtrinsics(const Mat& _imagePoints, const Mat& _objectPoints, const IntrinsicParams& param, Mat& omckk, Mat& Tckk)
1377 {
1378 CV_Assert(!_objectPoints.empty() && _objectPoints.type() == CV_64FC3);
1379 CV_Assert(!_imagePoints.empty() && _imagePoints.type() == CV_64FC2);
1380
1381 Mat imagePointsNormalized = NormalizePixels(_imagePoints, param).reshape(1).t();
1382 Mat objectPoints = _objectPoints.reshape(1).t();
1383 Mat objectPointsMean, covObjectPoints;
1384 Mat Rckk;
1385 int Np = imagePointsNormalized.cols;
1386 calcCovarMatrix(objectPoints, covObjectPoints, objectPointsMean, COVAR_NORMAL | COVAR_COLS);
1387 SVD svd(covObjectPoints);
1388 Mat R(svd.vt);
1389 if (norm(R(Rect(2, 0, 1, 2))) < 1e-6)
1390 R = Mat::eye(3,3, CV_64FC1);
1391 if (determinant(R) < 0)
1392 R = -R;
1393 Mat T = -R * objectPointsMean;
1394 Mat X_new = R * objectPoints + T * Mat::ones(1, Np, CV_64FC1);
1395 Mat H = ComputeHomography(imagePointsNormalized, X_new(Rect(0,0,X_new.cols,2)));
1396 double sc = .5 * (norm(H.col(0)) + norm(H.col(1)));
1397 H = H / sc;
1398 Mat u1 = H.col(0).clone();
1399 double norm_u1 = norm(u1);
1400 CV_Assert(fabs(norm_u1) > 0);
1401 u1 = u1 / norm_u1;
1402 Mat u2 = H.col(1).clone() - u1.dot(H.col(1).clone()) * u1;
1403 double norm_u2 = norm(u2);
1404 CV_Assert(fabs(norm_u2) > 0);
1405 u2 = u2 / norm_u2;
1406 Mat u3 = u1.cross(u2);
1407 Mat RRR;
1408 hconcat(u1, u2, RRR);
1409 hconcat(RRR, u3, RRR);
1410 Rodrigues(RRR, omckk);
1411 Rodrigues(omckk, Rckk);
1412 Tckk = H.col(2).clone();
1413 Tckk = Tckk + Rckk * T;
1414 Rckk = Rckk * R;
1415 Rodrigues(Rckk, omckk);
1416 }
1417
CalibrateExtrinsics(InputArrayOfArrays objectPoints,InputArrayOfArrays imagePoints,const IntrinsicParams & param,const int check_cond,const double thresh_cond,InputOutputArray omc,InputOutputArray Tc)1418 void cv::internal::CalibrateExtrinsics(InputArrayOfArrays objectPoints, InputArrayOfArrays imagePoints,
1419 const IntrinsicParams& param, const int check_cond,
1420 const double thresh_cond, InputOutputArray omc, InputOutputArray Tc)
1421 {
1422 CV_Assert(!objectPoints.empty() && (objectPoints.type() == CV_32FC3 || objectPoints.type() == CV_64FC3));
1423 CV_Assert(!imagePoints.empty() && (imagePoints.type() == CV_32FC2 || imagePoints.type() == CV_64FC2));
1424 CV_Assert(omc.type() == CV_64FC3 || Tc.type() == CV_64FC3);
1425
1426 if (omc.empty()) omc.create(1, (int)objectPoints.total(), CV_64FC3);
1427 if (Tc.empty()) Tc.create(1, (int)objectPoints.total(), CV_64FC3);
1428
1429 const int maxIter = 20;
1430
1431 for(int image_idx = 0; image_idx < (int)imagePoints.total(); ++image_idx)
1432 {
1433 Mat omckk, Tckk, JJ_kk;
1434 Mat image, object;
1435
1436 objectPoints.getMat(image_idx).convertTo(object, CV_64FC3);
1437 imagePoints.getMat (image_idx).convertTo(image, CV_64FC2);
1438
1439 bool imT = image.rows < image.cols;
1440 bool obT = object.rows < object.cols;
1441
1442 InitExtrinsics(imT ? image.t() : image, obT ? object.t() : object, param, omckk, Tckk);
1443
1444 ComputeExtrinsicRefine(!imT ? image.t() : image, !obT ? object.t() : object, omckk, Tckk, JJ_kk, maxIter, param, thresh_cond);
1445 if (check_cond)
1446 {
1447 SVD svd(JJ_kk, SVD::NO_UV);
1448 if(svd.w.at<double>(0) / svd.w.at<double>((int)svd.w.total() - 1) > thresh_cond )
1449 CV_Error( cv::Error::StsInternal, format("CALIB_CHECK_COND - Ill-conditioned matrix for input array %d",image_idx));
1450 }
1451 omckk.reshape(3,1).copyTo(omc.getMat().col(image_idx));
1452 Tckk.reshape(3,1).copyTo(Tc.getMat().col(image_idx));
1453 }
1454 }
1455
ComputeJacobians(InputArrayOfArrays objectPoints,InputArrayOfArrays imagePoints,const IntrinsicParams & param,InputArray omc,InputArray Tc,const int & check_cond,const double & thresh_cond,Mat & JJ2,Mat & ex3)1456 void cv::internal::ComputeJacobians(InputArrayOfArrays objectPoints, InputArrayOfArrays imagePoints,
1457 const IntrinsicParams& param, InputArray omc, InputArray Tc,
1458 const int& check_cond, const double& thresh_cond, Mat& JJ2, Mat& ex3)
1459 {
1460 CV_Assert(!objectPoints.empty() && (objectPoints.type() == CV_32FC3 || objectPoints.type() == CV_64FC3));
1461 CV_Assert(!imagePoints.empty() && (imagePoints.type() == CV_32FC2 || imagePoints.type() == CV_64FC2));
1462
1463 CV_Assert(!omc.empty() && omc.type() == CV_64FC3);
1464 CV_Assert(!Tc.empty() && Tc.type() == CV_64FC3);
1465
1466 int n = (int)objectPoints.total();
1467
1468 JJ2 = Mat::zeros(9 + 6 * n, 9 + 6 * n, CV_64FC1);
1469 ex3 = Mat::zeros(9 + 6 * n, 1, CV_64FC1 );
1470
1471 for (int image_idx = 0; image_idx < n; ++image_idx)
1472 {
1473 Mat image, object;
1474 objectPoints.getMat(image_idx).convertTo(object, CV_64FC3);
1475 imagePoints.getMat (image_idx).convertTo(image, CV_64FC2);
1476
1477 bool imT = image.rows < image.cols;
1478 Mat om(omc.getMat().col(image_idx)), T(Tc.getMat().col(image_idx));
1479
1480 std::vector<Point2d> x;
1481 Mat jacobians;
1482 projectPoints(object, x, om, T, param, jacobians);
1483 Mat exkk = (imT ? image.t() : image) - Mat(x);
1484
1485 Mat A(jacobians.rows, 9, CV_64FC1);
1486 jacobians.colRange(0, 4).copyTo(A.colRange(0, 4));
1487 jacobians.col(14).copyTo(A.col(4));
1488 jacobians.colRange(4, 8).copyTo(A.colRange(5, 9));
1489
1490 A = A.t();
1491
1492 Mat B = jacobians.colRange(8, 14).clone();
1493 B = B.t();
1494
1495 JJ2(Rect(0, 0, 9, 9)) += A * A.t();
1496 JJ2(Rect(9 + 6 * image_idx, 9 + 6 * image_idx, 6, 6)) = B * B.t();
1497
1498 JJ2(Rect(9 + 6 * image_idx, 0, 6, 9)) = A * B.t();
1499 JJ2(Rect(0, 9 + 6 * image_idx, 9, 6)) = JJ2(Rect(9 + 6 * image_idx, 0, 6, 9)).t();
1500
1501 ex3.rowRange(0, 9) += A * exkk.reshape(1, 2 * exkk.rows);
1502 ex3.rowRange(9 + 6 * image_idx, 9 + 6 * (image_idx + 1)) = B * exkk.reshape(1, 2 * exkk.rows);
1503
1504 if (check_cond)
1505 {
1506 Mat JJ_kk = B.t();
1507 SVD svd(JJ_kk, SVD::NO_UV);
1508 CV_Assert(svd.w.at<double>(0) / svd.w.at<double>(svd.w.rows - 1) < thresh_cond);
1509 }
1510 }
1511
1512 std::vector<uchar> idxs(param.isEstimate);
1513 idxs.insert(idxs.end(), 6 * n, 1);
1514
1515 subMatrix(JJ2, JJ2, idxs, idxs);
1516 subMatrix(ex3, ex3, std::vector<uchar>(1, 1), idxs);
1517 }
1518
EstimateUncertainties(InputArrayOfArrays objectPoints,InputArrayOfArrays imagePoints,const IntrinsicParams & params,InputArray omc,InputArray Tc,IntrinsicParams & errors,Vec2d & std_err,double thresh_cond,int check_cond,double & rms)1519 void cv::internal::EstimateUncertainties(InputArrayOfArrays objectPoints, InputArrayOfArrays imagePoints,
1520 const IntrinsicParams& params, InputArray omc, InputArray Tc,
1521 IntrinsicParams& errors, Vec2d& std_err, double thresh_cond, int check_cond, double& rms)
1522 {
1523 CV_Assert(!objectPoints.empty() && (objectPoints.type() == CV_32FC3 || objectPoints.type() == CV_64FC3));
1524 CV_Assert(!imagePoints.empty() && (imagePoints.type() == CV_32FC2 || imagePoints.type() == CV_64FC2));
1525
1526 CV_Assert(!omc.empty() && omc.type() == CV_64FC3);
1527 CV_Assert(!Tc.empty() && Tc.type() == CV_64FC3);
1528
1529 int total_ex = 0;
1530 for (int image_idx = 0; image_idx < (int)objectPoints.total(); ++image_idx)
1531 {
1532 total_ex += (int)objectPoints.getMat(image_idx).total();
1533 }
1534 Mat ex(total_ex, 1, CV_64FC2);
1535 int insert_idx = 0;
1536 for (int image_idx = 0; image_idx < (int)objectPoints.total(); ++image_idx)
1537 {
1538 Mat image, object;
1539 objectPoints.getMat(image_idx).convertTo(object, CV_64FC3);
1540 imagePoints.getMat (image_idx).convertTo(image, CV_64FC2);
1541
1542 bool imT = image.rows < image.cols;
1543
1544 Mat om(omc.getMat().col(image_idx)), T(Tc.getMat().col(image_idx));
1545
1546 std::vector<Point2d> x;
1547 projectPoints(object, x, om, T, params, noArray());
1548 Mat ex_ = (imT ? image.t() : image) - Mat(x);
1549 ex_.copyTo(ex.rowRange(insert_idx, insert_idx + ex_.rows));
1550 insert_idx += ex_.rows;
1551 }
1552
1553 meanStdDev(ex, noArray(), std_err);
1554 std_err *= sqrt((double)ex.total()/((double)ex.total() - 1.0));
1555
1556 Vec<double, 1> sigma_x;
1557 meanStdDev(ex.reshape(1, 1), noArray(), sigma_x);
1558 sigma_x *= sqrt(2.0 * (double)ex.total()/(2.0 * (double)ex.total() - 1.0));
1559
1560 Mat JJ2, ex3;
1561 ComputeJacobians(objectPoints, imagePoints, params, omc, Tc, check_cond, thresh_cond, JJ2, ex3);
1562
1563 sqrt(JJ2.inv(), JJ2);
1564
1565 errors = 3 * sigma_x(0) * JJ2.diag();
1566 rms = sqrt(norm(ex, NORM_L2SQR)/ex.total());
1567 }
1568
dAB(InputArray A,InputArray B,OutputArray dABdA,OutputArray dABdB)1569 void cv::internal::dAB(InputArray A, InputArray B, OutputArray dABdA, OutputArray dABdB)
1570 {
1571 CV_Assert(A.getMat().cols == B.getMat().rows);
1572 CV_Assert(A.type() == CV_64FC1 && B.type() == CV_64FC1);
1573
1574 int p = A.getMat().rows;
1575 int n = A.getMat().cols;
1576 int q = B.getMat().cols;
1577
1578 dABdA.create(p * q, p * n, CV_64FC1);
1579 dABdB.create(p * q, q * n, CV_64FC1);
1580
1581 dABdA.getMat() = Mat::zeros(p * q, p * n, CV_64FC1);
1582 dABdB.getMat() = Mat::zeros(p * q, q * n, CV_64FC1);
1583
1584 for (int i = 0; i < q; ++i)
1585 {
1586 for (int j = 0; j < p; ++j)
1587 {
1588 int ij = j + i * p;
1589 for (int k = 0; k < n; ++k)
1590 {
1591 int kj = j + k * p;
1592 dABdA.getMat().at<double>(ij, kj) = B.getMat().at<double>(k, i);
1593 }
1594 }
1595 }
1596
1597 for (int i = 0; i < q; ++i)
1598 {
1599 A.getMat().copyTo(dABdB.getMat().rowRange(i * p, i * p + p).colRange(i * n, i * n + n));
1600 }
1601 }
1602
JRodriguesMatlab(const Mat & src,Mat & dst)1603 void cv::internal::JRodriguesMatlab(const Mat& src, Mat& dst)
1604 {
1605 Mat tmp(src.cols, src.rows, src.type());
1606 if (src.rows == 9)
1607 {
1608 Mat(src.row(0).t()).copyTo(tmp.col(0));
1609 Mat(src.row(1).t()).copyTo(tmp.col(3));
1610 Mat(src.row(2).t()).copyTo(tmp.col(6));
1611 Mat(src.row(3).t()).copyTo(tmp.col(1));
1612 Mat(src.row(4).t()).copyTo(tmp.col(4));
1613 Mat(src.row(5).t()).copyTo(tmp.col(7));
1614 Mat(src.row(6).t()).copyTo(tmp.col(2));
1615 Mat(src.row(7).t()).copyTo(tmp.col(5));
1616 Mat(src.row(8).t()).copyTo(tmp.col(8));
1617 }
1618 else
1619 {
1620 Mat(src.col(0).t()).copyTo(tmp.row(0));
1621 Mat(src.col(1).t()).copyTo(tmp.row(3));
1622 Mat(src.col(2).t()).copyTo(tmp.row(6));
1623 Mat(src.col(3).t()).copyTo(tmp.row(1));
1624 Mat(src.col(4).t()).copyTo(tmp.row(4));
1625 Mat(src.col(5).t()).copyTo(tmp.row(7));
1626 Mat(src.col(6).t()).copyTo(tmp.row(2));
1627 Mat(src.col(7).t()).copyTo(tmp.row(5));
1628 Mat(src.col(8).t()).copyTo(tmp.row(8));
1629 }
1630 dst = tmp.clone();
1631 }
1632
compose_motion(InputArray _om1,InputArray _T1,InputArray _om2,InputArray _T2,Mat & om3,Mat & T3,Mat & dom3dom1,Mat & dom3dT1,Mat & dom3dom2,Mat & dom3dT2,Mat & dT3dom1,Mat & dT3dT1,Mat & dT3dom2,Mat & dT3dT2)1633 void cv::internal::compose_motion(InputArray _om1, InputArray _T1, InputArray _om2, InputArray _T2,
1634 Mat& om3, Mat& T3, Mat& dom3dom1, Mat& dom3dT1, Mat& dom3dom2,
1635 Mat& dom3dT2, Mat& dT3dom1, Mat& dT3dT1, Mat& dT3dom2, Mat& dT3dT2)
1636 {
1637 Mat om1 = _om1.getMat();
1638 Mat om2 = _om2.getMat();
1639 Mat T1 = _T1.getMat().reshape(1, 3);
1640 Mat T2 = _T2.getMat().reshape(1, 3);
1641
1642 //% Rotations:
1643 Mat R1, R2, R3, dR1dom1(9, 3, CV_64FC1), dR2dom2;
1644 Rodrigues(om1, R1, dR1dom1);
1645 Rodrigues(om2, R2, dR2dom2);
1646 JRodriguesMatlab(dR1dom1, dR1dom1);
1647 JRodriguesMatlab(dR2dom2, dR2dom2);
1648 R3 = R2 * R1;
1649 Mat dR3dR2, dR3dR1;
1650 dAB(R2, R1, dR3dR2, dR3dR1);
1651 Mat dom3dR3;
1652 Rodrigues(R3, om3, dom3dR3);
1653 JRodriguesMatlab(dom3dR3, dom3dR3);
1654 dom3dom1 = dom3dR3 * dR3dR1 * dR1dom1;
1655 dom3dom2 = dom3dR3 * dR3dR2 * dR2dom2;
1656 dom3dT1 = Mat::zeros(3, 3, CV_64FC1);
1657 dom3dT2 = Mat::zeros(3, 3, CV_64FC1);
1658
1659 //% Translations:
1660 Mat T3t = R2 * T1;
1661 Mat dT3tdR2, dT3tdT1;
1662 dAB(R2, T1, dT3tdR2, dT3tdT1);
1663 Mat dT3tdom2 = dT3tdR2 * dR2dom2;
1664 T3 = T3t + T2;
1665 dT3dT1 = dT3tdT1;
1666 dT3dT2 = Mat::eye(3, 3, CV_64FC1);
1667 dT3dom2 = dT3tdom2;
1668 dT3dom1 = Mat::zeros(3, 3, CV_64FC1);
1669 }
1670
median(const Mat & row)1671 double cv::internal::median(const Mat& row)
1672 {
1673 CV_Assert(row.type() == CV_64FC1);
1674 CV_Assert(!row.empty() && row.rows == 1);
1675 Mat tmp = row.clone();
1676 sort(tmp, tmp, 0);
1677 if ((int)tmp.total() % 2) return tmp.at<double>((int)tmp.total() / 2);
1678 else return 0.5 *(tmp.at<double>((int)tmp.total() / 2) + tmp.at<double>((int)tmp.total() / 2 - 1));
1679 }
1680
median3d(InputArray m)1681 cv::Vec3d cv::internal::median3d(InputArray m)
1682 {
1683 CV_Assert(m.depth() == CV_64F && m.getMat().rows == 1);
1684 Mat M = Mat(m.getMat().t()).reshape(1).t();
1685 return Vec3d(median(M.row(0)), median(M.row(1)), median(M.row(2)));
1686 }
1687