-
Notifications
You must be signed in to change notification settings - Fork 0
/
fishmodel.cpp
1529 lines (1213 loc) · 58.1 KB
/
fishmodel.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
#include "fishmodel.h"
#include "ellipse_detect.h"
#include "config.h"
#include <opencv2/video/tracking.hpp>
extern cv::Mat frameDebugC;
extern cv::Size gszTemplateImg;
extern cv::Point gptTail,gptHead;
//extern double eyeStepIncrement;
//extern int gFishTailSpineSegmentLength;
//extern int gFitTailIntensityScanAngleDeg;
//extern const int gcFishContourSize; //Fixed number of fish Contour Points
//extern double gTemplateMatchThreshold;
fishModel::fishModel()
{
bNewModel = true;
//stepUpdate = 1.0; // with fast rate and slow down with updates
bearingAngle = 0.0f;
Delta_bearingAngle = 0.0f; //last Change In bearing
lastTailFitError = 0.0;
matchScore = 0.0;
nFailedEyeDetectionCount = 0;
inactiveFrames = 0;
matchScore = 0;
//coreTriangle.push_back(cv::Point());
//coreTriangle.push_back(cv::Point());
//coreTriangle.push_back(cv::Point());
this->mouthPoint.x = 0;
this->mouthPoint.y = 0;
//this->leftEyeHull.clear();
//this->rightEyeHull.clear();
this->ID = 0;
zTrack.id = this->ID;
zTrack.colour = CV_RGB(255,0,0);
leftEyeTheta = 0; //In Degrees - A Value that looks wrong to show its not initialized
rightEyeTheta = 0; //In Degrees
c_spineSegL = gTrackerState.gFishTailSpineSegmentLength;
//mState = cv::Mat::zeros(stateSize,1, type);
}
/////deprecated tracks And Blobs Here - To be removed
//fishModel::fishModel(cvb::CvTrack* track,cvb::CvBlob* blob):fishModel()
//{
// this->ID = track->id;
// this->blobLabel = track->label;
// //this->track = track; //Copy Pointer
// this->bearingRads = cvb::cvAngle(blob);
// this->coreTriangle[2].x = track->centroid.x;
// this->coreTriangle[2].y = track->centroid.y;
// templateScore = 0;
// this->resetSpine();
//}
fishModel::fishModel(zftblob blob,int bestTemplateOrientation,cv::Point ptTemplateCenter):fishModel()
{
//stepUpdate = 1.0; // with fast rate and slow down with updates
inactiveFrames = 0;
this->ID = blob.hash() ;
//this->blobLabel = blob.hash();
zTrack.id = this->ID;
this->zfishBlob = blob; //Copy Localy
//this->track = NULL;
this->bearingRads = (float)bestTemplateOrientation*CV_PI/180.0;
this->bearingAngle = (float)bestTemplateOrientation;
this->ptRotCentre = ptTemplateCenter;
zTrack.centroid = ptTemplateCenter;
matchScore = 0;
this->resetSpine();
qDebug() << "<<KF Init>>";
// >>>> KF State Initialization
// intialization of KF...
KF.init(stateSize, measSize, contrSize, type);
// KF State Vectors //
mMeasurement = cv::Mat::zeros(measSize,1,type);
mState = cv::Mat::zeros(stateSize,1,type);
cv::setIdentity(KF.errorCovPre, Scalar::all(1e-2f));
cv::setIdentity(KF.errorCovPost, Scalar::all(1e-3f)); // default is 0, for smoothing try 0.1
// [x,y,v_x,v_y,angle,angle_v]
//qDebug() << "<<KF Set State>>";
mState.at<float>(0) = (float)ptTemplateCenter.x; //X
mState.at<float>(1) = (float)ptTemplateCenter.y; //Y
mState.at<float>(2) = 0.0f; //speed X
mState.at<float>(3) = 0.0f; //speed Y
mState.at<float>(4) = 0.0f; //Angle Diff // (float)bestTemplateOrientation;// (Deg)
mState.at<float>(5) = 0.0f; //Accel V Angle (Deg)
mState.at<float>(6) = 0.0f; //Not Used
mState.at<float>(7) = 10.0f;// Left Eye
mState.at<float>(8) = -10.0f; //Right Eye
// declare an array of floats to feed into Kalman Filter Transition Matrix, also known as State Transition Model
// Transition State Matrix A [x,y,v_x,v_y,angle,angle_v]
// Note: set dT at each processing step!
// [ 1 0 dT 0 0 0 ]
// [ 0 1 0 dT 0 0 ]
// [ 0 0 1 0 0 0 ]
// [ 0 0 0 1 0 0 ]
// [ 0 0 0 0 1 dT ]
// [ 0 0 0 0 0 1 ]
//qDebug() << "<<KF Init Transition M>>";
KF.transitionMatrix = cv::Mat::zeros(measSize, stateSize, type);
cv::setIdentity(KF.transitionMatrix,cv::Scalar::all(1.0f));
// Measure Matrix H [z_x, z_y, angle]
// [ 1 0 0 0 0 0 ]
// [ 0 1 0 0 0 0 ]
// [ 0 0 0 0 1 0 ]
KF.measurementMatrix = cv::Mat::zeros(measSize, stateSize, CV_32FC1);
cv::setIdentity(KF.measurementMatrix,cv::Scalar::all(1.0f));
mMeasurement.at<float>(0) = ptTemplateCenter.x;
mMeasurement.at<float>(1) = ptTemplateCenter.y;
mMeasurement.at<float>(2) = 0.0f; //speed X
mMeasurement.at<float>(3) = 0.0f; //speed Y
mMeasurement.at<float>(4) = 0.0f;// No change in angle initiallythis->bearingAngle;
mMeasurement.at<float>(5) = 0.0f; //V Angle Accell
mMeasurement.at<float>(6) = 0.0f; //Not Used
mMeasurement.at<float>(7) = 10.0f;//this->leftEye.getEyeAngle();
mMeasurement.at<float>(8) = -10.0f;//this->rightEye.getEyeAngle();
//KF.measurementMatrix.at<float>(0) = 1.0f;
//KF.measurementMatrix.at<float>(7) = 1.0f;
//KF.measurementMatrix.at<float>(16) = 1.0f;
// // Process Noise Covariance Matrix Q [E_x,E_y, E_v_x,E_v_y ,E_angle,Eangle_v]
// // [ Ex 0 0 0 0 0 0 0]
// // [ 0 Ey 0 0 0 0 0 0]
// // [ 0 0 Ev_x 0 0 0 0 0]
// // [ 0 0 0 Ev_y 0 0 0 0]
// // [ 0 0 0 0 Ea 0 0 0]
// // [ 0 0 0 0 0 Ea_v0 0]
// [ 0 0 0 0 0 0 lE 0]
// [ 0 0 0 0 0 0 0 rE]
// Setting processNoise Is critical To achieve Smoothing without excessive lagging when fish rapidly moves. - I found 1e-2 works well, but lags on turns, which are often rapid
cv::setIdentity(KF.processNoiseCov, cv::Scalar(1e-2)); // default is 1, for smoothing try 0.0001
//Maybe Noise Suppression too high at 1e-3 , introduces lag in position
// KF.processNoiseCov.at<float>(0,0) = 1e-2;
// KF.processNoiseCov.at<float>(1,1) = 1e-2;
// KF.processNoiseCov.at<float>(2,2) = 1e-3;
// KF.processNoiseCov.at<float>(3,3) = 1e-3;
KF.processNoiseCov.at<float>(4,4) = 1.2f; //Angle Change between frames
KF.processNoiseCov.at<float>(5,5) = 1.2f; //Angular Accell (V of Diff)
// KF.processNoiseCov.at<float>(4,5) = 1e-4f; //Angular Diff to Angle Accell
// KF.processNoiseCov.at<float>(6,6) = 0.0f; //Not Used
KF.processNoiseCov.at<float>(7,7) = 1e-2f; //Left Eye
KF.processNoiseCov.at<float>(8,8) = 1e-2f; //Right Eye
//KF.processNoiseCov.at<float>(35) = 1e-1f;
// Measures Noise Covariance Matrix R - Set high/low so Filter Follows Measurement more closely
cv::setIdentity(KF.measurementNoiseCov, cv::Scalar(2e-2f)); // default is 1, increasing should smooth but I get erratic behaviour
//KF.measurementNoiseCov.at<float>(2,2) = 1e-3f; //Angular Diff (Speed)- per frame
//KF.measurementNoiseCov.at<float>(3,3) = 1e-3f; //Angle Accell
//KF.measurementNoiseCov.at<float>(1,2) = 1e-2f; // Y speed X pos
//KF.measurementNoiseCov.at<float>(0,3) = 1e-2f; //X Speed - X
//KF.measurementNoiseCov.at<float>(2,4) = 1e-5f; //Y Speed - Y
KF.measurementNoiseCov.at<float>(4,4) = 1e-5f; //Angular Diff (Speed)- per frame
KF.measurementNoiseCov.at<float>(5,5) = 1e-5f; //Angle Accell
//KF.measurementNoiseCov.at<float>(4,5) = 1e-3f; //1e-1f; //Angular V_speed - Accell Covar
//KF.measurementNoiseCov.at<float>(5,6) = 0;//1e-4f;//1e-1f; //Angular Accelleration-
KF.measurementNoiseCov.at<float>(7,7) = 2e-1f; //Left Eye
KF.measurementNoiseCov.at<float>(8,8) = 2e-1f; //Right Eye
KF.statePre = mState;
KF.statePost = mState.clone();
//qDebug() << "Fish Model Construct.";
}
fishModel::~fishModel()
{
//Clear The Vectors Contained
this->zTrack.pointStack.clear();
this->zTrack.pointStack.shrink_to_fit();
this->zTrack.pointStackRender.clear();
// this->zTrack.pointStackRender.shrink_to_fit();
}
float fishModel::leftEyeAngle()
{
// if (this->leftEyeRect.size.width < this->leftEyeRect.size.height)
// return (this->leftEyeRect.angle-90.0)*CV_PI/180.0;
// else
// {
// return (this->leftEyeRect.angle)*CV_PI/180.0;
// }
//These Values Are Kalman Filtered
return leftEyeTheta; //leftEye.rectEllipse.angle;
}
/// \brief return (corrected for leading edge from horizontal line -Pi ... +Pi) Rectangle angle in Radians
float fishModel::rightEyeAngle()
{
//if (this->rightEyeRect.size.width < this->rightEyeRect.size.height)
// return (this->rightEyeRect.angle-90.0)*CV_PI/180.0;
//else
// {
// return (this->rightEyeRect.angle)*CV_PI/180.0;
// }
//These Values Are Kalman Filtered
return rightEyeTheta; //leftEye.rectEllipse.angle;
}
///
/// \brief fishModel::resetSpine make a straight Spline pointing towards Blobs Bearings Angle
///
void fishModel::resetSpine()
{
//Reset Legth
c_spineSegL = gTrackerState.gc_FishTailSpineSegmentLength_init;
this->spline.clear();
spline.reserve(c_spinePoints);
for (int i=0;i<c_spinePoints;i++)
{
splineKnotf sp;
//1st Spine Is in Opposite Direction of Movement and We align 0 degrees to be upwards (vertical axis)
//if (this->bearingRads > CV_PI)
if (this->bearingRads < 0.0f)
this->bearingRads += 2.0*CV_PI;
sp.angleRad = (this->bearingRads)-CV_PI ; // //Spine Looks In Opposite Direction
sp.spineSegLength = c_spineSegL; //Default Size
if (sp.angleRad < 0.0f)
sp.angleRad += 2.0*CV_PI;
//else
// sp.angleRad = (this->bearingRads)+CV_PI/2.0; //CV_PI/2 //Spine Looks In Opposite Direcyion
assert(!std::isnan(sp.angleRad && std::abs(sp.angleRad) <= 2.0f*CV_PI && (sp.angleRad) >= 0 ));
if (i==0)
{
sp.x = this->ptRotCentre.x;
sp.y = this->ptRotCentre.y;
}
else
{
//0 Degrees Is vertical Axis Looking Up
sp.x = spline[i-1].x + ((double)c_spineSegL)*sin(sp.angleRad);
sp.y = spline[i-1].y - ((double)c_spineSegL)*cos(sp.angleRad);
}
spline.push_back(sp); //Add Knot to spline
}
// // ///DEBUG
// for (int j=0; j<c_spinePoints;j++) //Rectangle Eye
// {
// cv::circle(frameDebugC,cv::Point(spline[j].x,spline[j].y),2,TRACKER_COLOURMAP[j],1);
// }
// drawSpine(frameDebugC);
// cv::waitKey(300);
}
///
/// \brief fishModel::getSpine Recalculates Point positions using Stored Knot Params (Angles)
/// /Calculates Spine Positions assumes initial point x0 y0 stored at 0 index of vector
void fishModel::calcSpline(t_fishspline& outspline)
{
//this->spline.clear();
//this->spline.push_back(this->coreTriangle[2]);
double dspineSegL = outspline[0].spineSegLength;
for (int i=1;i<c_spinePoints;i++)
{
outspline[i].x = outspline[i-1].x + (dspineSegL)*sin(outspline[i-1].angleRad);
outspline[i].y = outspline[i-1].y - (dspineSegL)*cos(outspline[i-1].angleRad);
outspline[i].spineSegLength = dspineSegL;
assert(!std::isnan(outspline[i].y) && !std::isnan(outspline[i].x));
assert(!std::isnan(outspline[i].angleRad));
}
}
///
/// \brief fishModel::getSpine VARIATION Make a Spine variation modifying a specific param value indicated by index
/// \param inspline vector Passed by value so the original is unchanged
/// \param outspline - The variational Spline
/// \return distance of variation in Config space
double fishModel::getdeltaSpline(t_fishspline inspline, t_fishspline& outspline,int idxparam,double sgn)
{
const double dAngleStep = -sgn*CV_PI/80.0;
double dvarSpineSeg = (double)inspline[0].spineSegLength;
double ret = 0.0;
outspline = inspline;
//If idxparam = 1,2 then we are varying initial Spline Point x0, y0 params
if (idxparam == 0)
{
ret = sgn*0.5;
outspline[0].x -= ret;
}else if (idxparam == 1)
{
ret = sgn*0.5;
outspline[0].y -= ret;
}// segment size,
else if (idxparam == 2)
{
// if (dvarSpineSeg < this->c_MaxSpineLengthLimit)
ret = sgn*0.3;
// else
// ret = sgn*0.0000001; //Stop
dvarSpineSeg += ret;
outspline[0].spineSegLength = dvarSpineSeg;
}
else //Index > 2 is spine Angles
{
outspline[idxparam-3].angleRad += dAngleStep;// Angle variation for this theta
ret = dAngleStep*dvarSpineSeg+cos(dAngleStep)*dvarSpineSeg; //rTheta
}
//Readjust xi,yi (In variational terms calc x_i = f(q1,q2,q3..), y_i = f(q1,q2,q3..)
calcSpline(outspline);
return ret; //Return Distance In Q space
}
///
/// \brief fishModel::getSplineParams - Returns vector of Cspace spine coordinates
/// \param inspline
/// \param outparams
/// \return
///
void fishModel::getSplineParams(t_fishspline& inspline,std::vector<double>& outparams)
{
outparams.clear();
outparams.reserve(c_spineParamCnt);
//Add x0 - yo
//outparams[0] = inspline[0].x;
//outparams[1] = inspline[0].y;
outparams.push_back(inspline[0].x);
outparams.push_back(inspline[0].y);
outparams.push_back(inspline[0].spineSegLength); //Assume same Length Across Spine, given by 1st knot
for (int i=0;i<(c_spinePoints);i++)
{
outparams.push_back(inspline[i].angleRad);
//outparams[i+2] = inspline[i].angleRad;
}
//outparams[0] = inspline[0].x;
//outparams[1] = inspline[0].y;
//for (int i=0;i<(c_spinePoints);i++)
// outparams[i+2] = inspline[i].angleRad;
}
/// \brief Modifies a Spline according to Cspace params
void fishModel::setSplineParams(t_fishspline& inspline,std::vector<double>& inparams)
{
double dvarSpineSegLength;
for (int i=0;i<(c_spineParamCnt);i++)
{
if (i==0)
inspline[0].x = inparams[0];
if (i==1)
inspline[0].y = inparams[1];
if (i==2) //Segment Size
{
dvarSpineSegLength = inparams[2];
//check if change in spineSegLength too sudden - And Reject //
if (abs(dvarSpineSegLength - inspline[0].spineSegLength) < 2 )
{
// Impose Limits
inspline[0].spineSegLength = min(c_MaxSpineLengthLimit,dvarSpineSegLength);
inspline[0].spineSegLength = max(c_MinSpineLengthLimit,(double)inspline[0].spineSegLength);
//assert(dvarSpineSegLength < 50);
}
}
if (i>2)
{
//Cap -Pi to Pi
inspline[i-3].angleRad = max(-CV_PI,min(CV_PI,(double)inparams[i])); //Param 3 is actually 1st spine knot's angle
//assert(inparams[i] >= -CV_PI && inparams[i] <= CV_PI);
}
}
//Readjust xi,yi (In variational terms calc x_i = f(q1,q2,q3..), y_i = f(q1,q2,q3..)
calcSpline(inspline);
}
float fishModel::vergenceAngle()
{
}
/// \brief Implements spine Curve function by combining piecewise elements
cv::Point2f fishModel::getPointAlongSpline(float z,t_fishspline& pspline)
{
//const float spineLength = this->c_spineSegL*pspline.size(); //The fitted Spine's lentgh is fixed
const float spineSegLength = pspline[0].spineSegLength;
const float spineLength = spineSegLength*pspline.size(); //The fitted Spine's lentgh is fixed
int idx = z/spineSegLength; //Find knot index which is contains point
double segLen = z - idx*spineSegLength; //Modulo Find length input var along a linear segment
if (idx > (pspline.size()-1)) //If Input Exceeds Spine Length
return cv::Point2f(pspline[pspline.size()-1].x,pspline[pspline.size()-1].y);
///Now construct point using Length along curve and return
cv::Point2f ptC;
ptC.x = pspline[idx].x + segLen*sin(pspline[idx].angleRad);
ptC.y = pspline[idx].y - segLen*cos(pspline[idx].angleRad);
return ptC;
}
///
/// \brief distancePointToSpline Finds Foot point - Important for measuring fit error between points and spline
/// Different Schemes exists - such as PDM, TDM - SDM - Start with Point Distance Radial Measure
/// \param pt
/// \param spline
/// \return
///
double fishModel::distancePointToSpline(cv::Point2f ptsrc,t_fishspline& pspline)
{
//const int spineLength = this->c_spineSegL*pspline.size(); //The fitted Spine's lentgh is fixed
const int spineLength = pspline[0].spineSegLength*pspline.size(); //The fitted Spine's varies
const double dCStep = 0.3; //Step size on when searching along Spine Curve for closest Point (Foot Point )tk
int idxNear = 0;
//Take Spine Point From Body / Set As foot point
cv::Point2f ptFoot = cv::Point2f(pspline[idxNear].x,pspline[idxNear].y);
//double distX = pow(ptsrc.x-ptFoot.x,2);
//double distY = pow(ptsrc.y-ptFoot.y,2);
double mindist;// = sqrt(distX + distY);
mindist= cv::norm(ptsrc-ptFoot); //Start from Foot / Body Point look for point closer than this
double dist = mindist;
float fScanC = 0.0;
///Find Closest point on Curve to POint
/// \todo this is a crude/naive search -ok for small spine lengths
/// Best to improve by calculating/estimating FootPoint projection
while (fScanC < spineLength)
{
cv::Point2f ptTest = getPointAlongSpline(fScanC,pspline);
//distX = pow(ptsrc.x-ptTest.x,2);
//distY = pow(ptsrc.y-ptTest.y,2);
//dist = sqrt(distX + distY);
dist= cv::norm(ptsrc-ptTest);
//Check if distance minimized -
if (dist < mindist)
{
mindist = dist;
ptFoot = ptTest;
}
fScanC += dCStep; //Move Along Curve
}
#ifdef _ZTFDEBUG_
//Show Foot Points
cv::circle(frameDebugC,ptFoot,1,CV_RGB(10,10,255),1);
#endif
return mindist;
}
/// \brief Uses detected ellipsoids to set fish's eye model state / using an incremental update
///\return total Score for fit
int fishModel::updateEyeMeasurement(tEllipsoids& vLeftEll,tEllipsoids& vRightEll)
{
int retPerfScore = 0;
double fleftEyeTheta = 0.0f;
//int ileftEyeSamples = 0;
double frightEyeTheta = 0.0f;
//int irightEyeSamples = 0;
// If we are stuck on same frame then estimate the unbiased empirical mean angle for each eye
// use an incremental mean calculation
// if (uiFrameIterations > 1)
// stepUpdate = 1.0/std::min(200.0, (double)uiFrameIterations);
tDetectedEllipsoid mleftEye(vLeftEll);
tDetectedEllipsoid mrightEye(vRightEll);
fleftEyeTheta = mleftEye.getEyeAngle();
frightEyeTheta = mrightEye.getEyeAngle();
//Incremental Update
if (std::isnan(fleftEyeTheta) )
this->nFailedEyeDetectionCount++; //Do not Update Measurement
else
{
mMeasurement.at<float>(7) = fleftEyeTheta;//this->leftEyeTheta + stepUpdate*(fleftEyeTheta - this->leftEyeTheta );
this->lastLeftEyeMeasured = mleftEye; // Copied Here to update axis Position, But Angle is corrected via Kalman Filtering
this->lastLeftEyeMeasured.rectEllipse.angle = this->leftEyeTheta;
}
if (std::isnan(frightEyeTheta) )
this->nFailedEyeDetectionCount++;
else{
//this->rightEyeTheta = this->rightEyeTheta + stepUpdate*(frightEyeTheta - this->rightEyeTheta );
mMeasurement.at<float>(8) = frightEyeTheta;//this->leftEyeTheta + stepUpdate*(fleftEyeTheta - this->leftEyeTheta );
this->lastRightEyeMeasured = mrightEye; // Copied Here to update axis Position, But Angle is corrected via Kalman Filtering
this->lastRightEyeMeasured.rectEllipse.angle = this->rightEyeTheta;
}
if (mleftEye.fitscore > 0 && mrightEye.fitscore > 0)
{
this->nFailedEyeDetectionCount = 0; // Reset Error Count
retPerfScore = this->lastLeftEyeMeasured.fitscore,this->lastRightEyeMeasured.fitscore;
}else //penalize
{
retPerfScore = (mleftEye.fitscore + mrightEye.fitscore)- 400;
}
//Reset Step size to default
// stepUpdate = gTrackerState.eyeStepIncrement;
return (retPerfScore);
// //tDetectedEllipsoid
// // Go through All detected ellipsoids,
// for (int i=0; i< vLeftEll.size(); i++)
// {
// // select ones are for left/right eye
// tDetectedEllipsoid Eye = vLeftEll[i];
// if (Eye.cLabel == 'L') //Left eye
// {
// fleftEyeTheta += Eye.getEyeAngle();
// ileftEyeSamples +=1;
// }
// // and obtain mean angle for left/right eye from set of detected ellipsoids.
// }
// for (int i=0; i< vRightEll.size(); i++)
// {
// tDetectedEllipsoid REye = vRightEll[i];
// frightEyeTheta += REye.getEyeAngle();
// irightEyeSamples +=1;
// this->leftEye.fitscore += REye.fitscore;
// }
//Get Mean sample angles
// if (ileftEyeSamples >0)
// {
// fleftEyeTheta = fleftEyeTheta/(float)ileftEyeSamples;
// this->leftEye.fitscore = this->leftEye.fitscore/(float)ileftEyeSamples;
// }else
// {
// this->nFailedEyeDetectionCount++;
// }
// if (irightEyeSamples >0)
// {
// frightEyeTheta = frightEyeTheta/(float)irightEyeSamples;
// this->rightEye.fitscore = this->rightEye.fitscore/(float)irightEyeSamples;
// }else
// {
// this->nFailedEyeDetectionCount++;
// }
// if (vell.size() > 0)
// {//Left Eye Detected First
// tDetectedEllipsoid lEye = vell.at(0); //L Eye Is pushed 1st
// // Update Internal Variable for Eye Angle //
// // Use an incremental/ recent average rule
// lEye.rectEllipse.angle = fleftEyeTheta;
// this->leftEye = lEye;
// if (lEye.fitscore > 50)
// this->leftEyeTheta = this->leftEyeTheta + stepUpdate*(fleftEyeTheta - this->leftEyeTheta );
// }else
// { //Set To Not detected - Do not update estimates - set score to 0
// //this->leftEye = tDetectedEllipsoid(cv::RotatedRect(),0);
// //this->leftEyeTheta = 180;
// this->leftEye.fitscore = 0;
// this->nFailedEyeDetectionCount++;
// }
// // ss.str(""); //Empty String
// if (vell.size() > 1)
// {
// tDetectedEllipsoid rEye = vell.at(1); //R Eye Is pushed 2nd
// frightEyeTheta = rEye.rectEllipse.angle - 90;
// //Fix Equivalent Angles To Range -50 +30
// if (frightEyeTheta < -90)
// frightEyeTheta = rEye.rectEllipse.angle+90;
// if (frightEyeTheta > 30)
// frightEyeTheta = rEye.rectEllipse.angle-90;
// rEye.rectEllipse.angle = frightEyeTheta;
// this->rightEye = rEye; //Save last fitted ellipsoid struct
// // Update Internal Variable for Eye Angle //
// // Use an incremental/ recent average rule
// if (rEye.fitscore > 50)
// this->rightEyeTheta = this->rightEyeTheta + stepUpdate*(frightEyeTheta - this->rightEyeTheta );
// }else
// { //Set To Not detected
// // ss << "R Eye Detection Error - Check Threshold";
// // window_main.LogEvent(QString::fromStdString(ss.str()));
// //this->rightEye = tDetectedEllipsoid(cv::RotatedRect(),0);
// //this->rightEyeTheta = 180;
// this->rightEye.fitscore = 0;
// this->nFailedEyeDetectionCount++;
// }
}
void fishModel::drawAnteriorBox(cv::Mat& frameScene, cv::Scalar colour=CV_RGB(00,00,255))
{
///Draw a Red Rotated Frame around Detected Body
cv::Point2f boundBoxPnts[4];
bodyRotBound.points(boundBoxPnts);
for (int j=0; j<4;j++) //Rectangle Body
cv::line(frameScene,boundBoxPnts[j],boundBoxPnts[(j+1)%4],colour,1,cv::LINE_8);
}
///
/// \brief fishModel::Update - Called On Every Frame Processed To Update Model State
/// The track point is set to the blob position and not the template centre
/// \param fblob
/// \param templatematchScore
/// \param Angle
/// \param bcentre
///
bool fishModel::stepPredict(unsigned int nFrame)
{
double dT = (double)(nFrame-nLastUpdateFrame);///((double)gTrackerState.gfVidfps+1.0)
//cout << "T:" << KF.transitionMatrix << endl;
if (!bNewModel) // First detection!
{ //Add Speed Contributions
KF.transitionMatrix.at<float>(0,2) = 1.0;
KF.transitionMatrix.at<float>(1,3) = 1.0;
KF.transitionMatrix.at<float>(4,5) = 1.0f;//1e-9f;// 0.01; //Angular Accell Feeds into Angle Diff (Speed)
KF.transitionMatrix.at<float>(5,6) = 0.0f; //1e-9f;//0.01; //Angular Speed Diff Feeds into Angular Speed
mState = KF.predict();
if (std::isnan(mState.at<float>(4)))
{
cerr << "[KalmanERROR] Sm:" << mState << endl;
KF.init(stateSize, measSize, contrSize, type); //Re Init
return(false);
}
// if (abs(mState.at<float>(4) - this->bearingAngle) > 20 )
// qDebug() << "KF Angle prediction error:" << this->bearingAngle << "->" << mState.at<float>(4);
//Integrate
//this->Delta_bearingAngle = mState.at<float>(4);
this->bearingAngle = (int)(zfishBlob.angle + mState.at<float>(4))%360; // Angle;
this->bearingRads = this->bearingAngle*CV_PI/180.0;
assert(!std::isnan(this->bearingAngle));
this->ptRotCentre = cv::Point2f(mState.at<float>(0), mState.at<float>(1)); //bcentre;
this->leftEyeTheta = mState.at<float>(7); // Eye Angle Left;
this->rightEyeTheta = mState.at<float>(8); // Eye Angle Right;
bPredictedPosition = true;
return (true);
}
return (false);
}
///
/// \brief fishModel::Update - Called On Every Frame Where Measurements from Blob are available - To Update Model State
/// The track point is set to the Kalman filtered blob state
///
/// \param fblob
/// \param matchScore - The fish blob's classifier score
/// \param Angle
/// \param bcentre
///
bool fishModel::updateState(zftblob* fblob, cv::Point2f bcentre,unsigned int nFrame,int SpineSegLength,int TemplRow, int TemplCol)
{
//assert(!std::isnan(Angle));
//Compare displacements to Last Measurements Not To Predicted Position In BearingAngle
// Note Blob Angles can flip 180 between frames so we need to take the closest to the current orientation
// \note Issue with Mod numbers Compass - Correcting towards from 355 to 0 the wrong way
float angleDisplacement;
float angleDisplacementA = getAngleDiff(zfishBlob.angle,this->bearingAngle);
float angleDisplacementB = getAngleDiff((int)(zfishBlob.angle+180)%360,this->bearingAngle);
//Choose the Displacement Closer to Current Angle (Fix Blob Noisy angle inversions)
angleDisplacement = (abs(angleDisplacementA) < abs(angleDisplacementB))?angleDisplacementA:angleDisplacementB;
//Angle = (abs(angleDisplacementA) < abs(angleDisplacementB))?Angle:(int)(Angle+180.0f)%360;
double stepDisplacement = cv::norm(bcentre - this->zTrack.centroid);
float dT = (float)(nFrame-nLastUpdateFrame);///((double)gTrackerState.gfVidfps+1.0)
if (bNewModel)
dT = 0;
else
if (angleDisplacement/dT > 150 ) //Correct Large Change - Blob Angle Flipped
{
qDebug() << "[W] Large angle change detected";
//return (false);
}
if (!gTrackerState.bDraggingTemplateCentre)
bUserDrag = false;
//qDebug() << "Fish-Update M.";
///\note mod angles cannot be KF tracked as transitions 0->360 are non linear - instead I track the change in angle and integrate
//Set to 1 frame minimum time step
KF.transitionMatrix.at<float>(0,2) = dT;
KF.transitionMatrix.at<float>(1,3) = dT;
KF.transitionMatrix.at<float>(4,5) = dT;//0.0f; // Angle Accell Feeds into Angle V
KF.transitionMatrix.at<float>(5,6) = 0.0f;//dT;//dT; //Angular V Diff Feeds into Angle V
//KF.transitionMatrix.at<float>(5,4) = 0;//dT;
mMeasurement.at<float>(0) = bcentre.x;
mMeasurement.at<float>(1) = bcentre.y;
mMeasurement.at<float>(2) = mMeasurement.at<float>(3) = mMeasurement.at<float>(4) = mMeasurement.at<float>(5) = 0.0f;
//mMeasurement.at<float>(7) = this->leftEye.getEyeAngle(); Updated Separetelly In UpdateEyeState
//mMeasurement.at<float>(8) = this->rightEye.getEyeAngle();
if (dT > 0){
stepDisplacement = stepDisplacement/dT;
angleDisplacement = angleDisplacement/dT;
//Add Speed as measured Blob speed (Do not involve Filter Predictions in Measured Speed)
mMeasurement.at<float>(2) = (bcentre.x-zfishBlob.pt.x)/dT;
mMeasurement.at<float>(3) = (bcentre.y-zfishBlob.pt.y)/dT; //Y speed;
mMeasurement.at<float>(4) = (float)angleDisplacement/dT;
mMeasurement.at<float>(5) = Delta_bearingAngle - angleDisplacement;//Angle - this->bearingAngle; //angleDisplacement; //min(1.0f,max(-1.0f,(float)(angleDisplacement)/1.0f)); //(geAngleDiff(zfishBlob.angle,Angle)); //Ang Speed
//mMeasurement.at<float>(6) = 0;//(floqat)(mState.at<float>(5) - (angleDisplacement))/2.0f; // min(0.1f,max(-0.1f,(float)(mState.at<float>(5) - (angleDisplacement))/2.0f)); //Ang Accelleration
}
//else
// mMeasurement.at<float>(2) = mMeasurement.at<float>(3) = mMeasurement.at<float>(5) = mMeasurement.at<float>(6) =0;
// >>>> Matrix A - Note: set dT at each processing step :
/// Kalman Update - Measurements From Blob //
/// generate measurement
//mMeasurement += KF.measurementMatrix*mState;
/// Kalman FILTER //
/// Re-Order - First adjust to measurement - then Predict
//Reject Updates That Are Beyond Bounds
// if (stepDisplacement > gTrackerState.gDisplacementLimitPerFrame ||
// angleDisplacement > gTrackerState.gAngleChangeLimitPerFrame){
// inactiveFrames++;
// }else{ //Measurement valid - C0nsume
mCorrected = KF.correct(mMeasurement); // Kalman Correction
//mCorrected.copyTo(mState);
// if (abs(mCorrected.at<float>(4) - Angle) > 20)
// qDebug() << "KF Angle Meas:" << Angle << " Error Pred:" << mState.at<float>(4) << " Corrected " << mCorrected.at<float>(4);
//Catch KALMAN error and skip update
if (std::isnan(mCorrected.at<float>(0)))
return(false);
inactiveFrames = 0;
// }
this->zTrack.id = ID;
this->matchScore = fblob->response;// templatematchScore;
this->Delta_bearingAngle = mCorrected.at<float>(4);
this->bearingAngle = zfishBlob.angle + mCorrected.at<float>(4); // Integrate Angle Change onto Bearing;
assert(!std::isnan(this->bearingAngle));
this->bearingRads = this->bearingAngle*CV_PI/180.0;
this->ptRotCentre = cv::Point2f(mCorrected.at<float>(0), mCorrected.at<float>(1)); //bcentre;
this->leftEyeTheta = mCorrected.at<float>(7); // Eye Angle Left;
this->rightEyeTheta = mCorrected.at<float>(8); // Eye Angle Right;
//Update The Eye Ellipsoids
this->lastLeftEyeMeasured.rectEllipse.angle = this->leftEyeTheta;
this->lastRightEyeMeasured.rectEllipse.angle = this->rightEyeTheta;
//this->leftEye = tDetectedEllipsoid(this->leftEye.rectEllipse,this->leftEye.fitscore);
//this->rightEye = tDetectedEllipsoid(this->rightEye.rectEllipse,this->rightEye.fitscore);
//Blob Position Is not FIltered
this->zfishBlob = *fblob;
//this->c_spineSegL = SpineSegLength;
this->zTrack.pointStack.push_back(this->ptRotCentre);
this->zTrack.addRenderPoint(this->ptRotCentre);
this->zTrack.effectiveDisplacement = cv::norm(this->ptRotCentre - this->zTrack.centroid);
this->zTrack.centroid = this->ptRotCentre;//fblob->pt; //Or Maybe bcentre
if (this->zTrack.effectiveDisplacement > gTrackerState.gDisplacementThreshold)
{
this->zTrack.active++;
}else {
this->zTrack.inactive++;
}
/// Update Template Box Bound
//int bestAngleinDeg = fish->bearingAngle;
cv::RotatedRect fishRotAnteriorBox(this->ptRotCentre,gTrackerState.gszTemplateImg ,this->bearingAngle); //fblob->pt
/// Save Anterior Bound
this->bodyRotBound = fishRotAnteriorBox;
this->idxTemplateCol = TemplCol;
this->idxTemplateRow = TemplRow;
//Set Spine Source to - Corrected Position
this->spline[0].x = this->ptRotCentre.x;
this->spline[0].y = this->ptRotCentre.y;
//this->spline[0].angleRad = this->bearingRads+CV_PI; //+180 Degrees so it looks in Opposite Direction
//Check if frame advanced
if (nLastUpdateFrame == nFrame)
uiFrameIterations++; //Increment count of calculation cycles that we are stuck on same frame
else
{
nLastUpdateFrame = nFrame; //Set Last Update To Current Frame
uiFrameIterations = 0;
}
bNewModel = false; //Flag THat this model Has been now positioned
bPredictedPosition = false; // position is based on corrected measurement
return(true);
}//End of UpdateState
/// \brief Revised fitSpineContour , V2- Faster as it focuses/iterates around the spine points not the contour points
/// and uses the OpenCV pointPolygonTest to measure point distance to contour.
/// \param contours_body Fish Body Contour
/// \param frameImg_grey
double fishModel::fitSpineToContour2(cv::Mat& frameImg_grey, std::vector<std::vector<cv::Point> >& contours_body,int idxInnerContour,int idxOuterContour)
{
static const int cntParam = this->c_spineParamCnt;
static const int cFitSpinePointsCount = ZTF_TAILSPINECOUNT;//
//Parameter Found By Experience for current size fish
///Param sfish model should contain initial spline curve (Hold Last Frame Position)
//Run Until Convergence Error is below threshold - Or Change is too small
///Compute Error terms for all data points/obtain local quadratic approx of fsd
//For each contour Point
this->contour = contours_body[idxOuterContour];
t_fishspline tmpspline = this->spline;
t_fishspline dsSpline; //Variational Spline
//Measure squared Distance error to closest Curve(spline) Point
//Add to total error
double dfitPtError_total = 10000.0;
double dfitPtError_total_last = 0.0;
double dDifffitPtError_total = 1000.0;
double dTemp = 1.0; //Anealling Temperature
/// \todo Optimize - Make Fish Contour Size Fixed - Then Allocate this as a buffer on the heap and reuse
double dJacobian[cFitSpinePointsCount][cntParam];//Vector of \nabla d for error functions
memset(dJacobian,0.0,cFitSpinePointsCount*(cntParam)*sizeof(double));
double dGradf[cntParam];//Vector of Grad F per param
memset(dGradf,0.0,cntParam*sizeof(double));
double dGradi[cntParam];//Vector of Grad Intensity per SPine POint param
memset(dGradi,0.0,cntParam*sizeof(double));
double dResiduals[cFitSpinePointsCount];//Vector of \nabla d for error functions
memset(dResiduals,0.0,cFitSpinePointsCount*sizeof(double));
int cntpass = 0;
int cntStuck = 0; //Number of Cycles solution has converge to an unnacceptable solution
int cntSolved = 0; //Number of Cycles Solution Is acceptable
double dVarScale = 1.0;
//Do A number of Passes Before Convergence //&& (dfitPtError_total/contour.size() > 8)
while (cntpass < gTrackerState.gMaxFitIterations && (cntStuck < 5) && (cntSolved < 3) )
{
//Converged But Error Is still Large per Countour Point Then Jolt
if (std::abs(dDifffitPtError_total) < 0.01 && dfitPtError_total/contour.size() > c_fitErrorPerContourPoint) //Time Out Convergece Count
{
cntStuck++;
dVarScale = -dVarScale*1.0; //*1.2
}
else
{
cntStuck = 0;
dVarScale = 1.0;
}
//Check For Ealy Convergence And Stop Early
if (std::abs(dDifffitPtError_total) <= c_fitErrorPerContourPoint) //Time Out Convergece Count
{
cntSolved++;
//dVarScale = dVarScale*0.93;
}
else
{
cntSolved = 0;
dVarScale = 1.0;
}
//Reset Grad INfo - Start Pass From Last Point
memset(dGradi,0.0,cntParam*sizeof(double));
memset(dGradf,0.0,cntParam*sizeof(double));
cntpass++;
///For Annealing
dTemp = (double)cntpass/gTrackerState.gMaxFitIterations;
//Prob Of Acceptance P = exp(-(sn-s)/T) >= drand
///
dfitPtError_total_last = dfitPtError_total;
dfitPtError_total = 0.0; //Reset
double dq,ds; //Variation In Space And Score Variation
//For Each Contour Point
///\todo invert the problem, go through each spine point and check against contour
for (uint i=1;i<cFitSpinePointsCount;i+=1) //For Each Data point make a row in Jacobian
{
//dResiduals[i] = distancePointToSpline((cv::Point2f)contour[i],tmpspline);
//Distance to closest Contour Edge, +ve inside, 0 on edge -ve outside
// Invert so minimum is at centre of contour
dResiduals[i] = -pointPolygonTest(contour, cv::Point2f(tmpspline[i].x,tmpspline[i].y), true );
//Add Extra Grad Info/Cost for small segLength / thus pulling to longer spine length
// dResiduals[i] -= (0.05)*c_MaxSpineLengthLimit/tmpspline[i].spineSegLength;
// Push to get Tail Pt Coincide with Last spine point
double distToTailTip = cv::norm( gptTail - cv::Point(tmpspline[tmpspline.size()-1].x,tmpspline[tmpspline.size()-1].y) );
dResiduals[i] += (0.1)*distToTailTip/cFitSpinePointsCount;
// double penalty = dResiduals[i]*0.10; //Calc Scaled Penalty
// for (int s=0;s<tmpspline.size();s++)
// {
// int pptTest = pointPolygonTest(contour, cv::Point2f(tmpspline[s].x,tmpspline[s].y), false );
// if (pptTest < 0 ) //if spine point is outside contour then increase residuals
// dResiduals[i] += penalty;
//}
dfitPtError_total +=dResiduals[i];
//Add Variation dx to each param and calc derivative
//Start from param idx 2 thus skipping the 1st point(root ) position and only do angle variations
for (int k=2;k < cntParam; k++)
{ /// \note using only +ve dx variations and not -dx - In this C space Ds magnitude should be symmetrical to dq anyway
dq = getdeltaSpline(tmpspline,dsSpline,k,dVarScale); //Return param variation
// dsSpline residual of variation spline
ds = -pointPolygonTest(contour, cv::Point2f(dsSpline[i].x,dsSpline[i].y), true );
{
dJacobian[i][k] = (ds-dResiduals[i])/(dq);