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, Willow Garage Inc., all rights reserved.
15// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
16// Third party copyrights are property of their respective owners.
17//
18// Redistribution and use in source and binary forms, with or without modification,
19// are permitted provided that the following conditions are met:
20//
21// * Redistribution's of source code must retain the above copyright notice,
22// this list of conditions and the following disclaimer.
23//
24// * Redistribution's in binary form must reproduce the above copyright notice,
25// this list of conditions and the following disclaimer in the documentation
26// and/or other materials provided with the distribution.
27//
28// * The name of the copyright holders may not be used to endorse or promote products
29// derived from this software without specific prior written permission.
30//
31// This software is provided by the copyright holders and contributors "as is" and
32// any express or implied warranties, including, but not limited to, the implied
33// warranties of merchantability and fitness for a particular purpose are disclaimed.
34// In no event shall the Intel Corporation or contributors be liable for any direct,
35// indirect, incidental, special, exemplary, or consequential damages
36// (including, but not limited to, procurement of substitute goods or services;
37// loss of use, data, or profits; or business interruption) however caused
38// and on any theory of liability, whether in contract, strict liability,
39// or tort (including negligence or otherwise) arising in any way out of
40// the use of this software, even if advised of the possibility of such damage.
41//
42//M*/
43
44#ifndef OPENCV_TRACKING_HPP
45#define OPENCV_TRACKING_HPP
46
47#include "opencv2/core.hpp"
48#include "opencv2/imgproc.hpp"
49#ifdef HAVE_OPENCV_DNN
50# include "opencv2/dnn.hpp"
51#endif
52
53namespace cv
54{
55
56//! @addtogroup video_track
57//! @{
58
59enum { OPTFLOW_USE_INITIAL_FLOW = 4,
60 OPTFLOW_LK_GET_MIN_EIGENVALS = 8,
61 OPTFLOW_FARNEBACK_GAUSSIAN = 256
62 };
63
64/** @brief Finds an object center, size, and orientation.
65
66@param probImage Back projection of the object histogram. See calcBackProject.
67@param window Initial search window.
68@param criteria Stop criteria for the underlying meanShift.
69returns
70(in old interfaces) Number of iterations CAMSHIFT took to converge
71The function implements the CAMSHIFT object tracking algorithm @cite Bradski98 . First, it finds an
72object center using meanShift and then adjusts the window size and finds the optimal rotation. The
73function returns the rotated rectangle structure that includes the object position, size, and
74orientation. The next position of the search window can be obtained with RotatedRect::boundingRect()
75
76See the OpenCV sample camshiftdemo.c that tracks colored objects.
77
78@note
79- (Python) A sample explaining the camshift tracking algorithm can be found at
80 opencv_source_code/samples/python/camshift.py
81 */
82CV_EXPORTS_W RotatedRect CamShift( InputArray probImage, CV_IN_OUT Rect& window,
83 TermCriteria criteria );
84/** @example samples/cpp/camshiftdemo.cpp
85An example using the mean-shift tracking algorithm
86*/
87
88/** @brief Finds an object on a back projection image.
89
90@param probImage Back projection of the object histogram. See calcBackProject for details.
91@param window Initial search window.
92@param criteria Stop criteria for the iterative search algorithm.
93returns
94: Number of iterations CAMSHIFT took to converge.
95The function implements the iterative object search algorithm. It takes the input back projection of
96an object and the initial position. The mass center in window of the back projection image is
97computed and the search window center shifts to the mass center. The procedure is repeated until the
98specified number of iterations criteria.maxCount is done or until the window center shifts by less
99than criteria.epsilon. The algorithm is used inside CamShift and, unlike CamShift , the search
100window size or orientation do not change during the search. You can simply pass the output of
101calcBackProject to this function. But better results can be obtained if you pre-filter the back
102projection and remove the noise. For example, you can do this by retrieving connected components
103with findContours , throwing away contours with small area ( contourArea ), and rendering the
104remaining contours with drawContours.
105
106 */
107CV_EXPORTS_W int meanShift( InputArray probImage, CV_IN_OUT Rect& window, TermCriteria criteria );
108
109/** @brief Constructs the image pyramid which can be passed to calcOpticalFlowPyrLK.
110
111@param img 8-bit input image.
112@param pyramid output pyramid.
113@param winSize window size of optical flow algorithm. Must be not less than winSize argument of
114calcOpticalFlowPyrLK. It is needed to calculate required padding for pyramid levels.
115@param maxLevel 0-based maximal pyramid level number.
116@param withDerivatives set to precompute gradients for the every pyramid level. If pyramid is
117constructed without the gradients then calcOpticalFlowPyrLK will calculate them internally.
118@param pyrBorder the border mode for pyramid layers.
119@param derivBorder the border mode for gradients.
120@param tryReuseInputImage put ROI of input image into the pyramid if possible. You can pass false
121to force data copying.
122@return number of levels in constructed pyramid. Can be less than maxLevel.
123 */
124CV_EXPORTS_W int buildOpticalFlowPyramid( InputArray img, OutputArrayOfArrays pyramid,
125 Size winSize, int maxLevel, bool withDerivatives = true,
126 int pyrBorder = BORDER_REFLECT_101,
127 int derivBorder = BORDER_CONSTANT,
128 bool tryReuseInputImage = true );
129
130/** @example samples/cpp/lkdemo.cpp
131An example using the Lucas-Kanade optical flow algorithm
132*/
133
134/** @brief Calculates an optical flow for a sparse feature set using the iterative Lucas-Kanade method with
135pyramids.
136
137@param prevImg first 8-bit input image or pyramid constructed by buildOpticalFlowPyramid.
138@param nextImg second input image or pyramid of the same size and the same type as prevImg.
139@param prevPts vector of 2D points for which the flow needs to be found; point coordinates must be
140single-precision floating-point numbers.
141@param nextPts output vector of 2D points (with single-precision floating-point coordinates)
142containing the calculated new positions of input features in the second image; when
143OPTFLOW_USE_INITIAL_FLOW flag is passed, the vector must have the same size as in the input.
144@param status output status vector (of unsigned chars); each element of the vector is set to 1 if
145the flow for the corresponding features has been found, otherwise, it is set to 0.
146@param err output vector of errors; each element of the vector is set to an error for the
147corresponding feature, type of the error measure can be set in flags parameter; if the flow wasn't
148found then the error is not defined (use the status parameter to find such cases).
149@param winSize size of the search window at each pyramid level.
150@param maxLevel 0-based maximal pyramid level number; if set to 0, pyramids are not used (single
151level), if set to 1, two levels are used, and so on; if pyramids are passed to input then
152algorithm will use as many levels as pyramids have but no more than maxLevel.
153@param criteria parameter, specifying the termination criteria of the iterative search algorithm
154(after the specified maximum number of iterations criteria.maxCount or when the search window
155moves by less than criteria.epsilon.
156@param flags operation flags:
157 - **OPTFLOW_USE_INITIAL_FLOW** uses initial estimations, stored in nextPts; if the flag is
158 not set, then prevPts is copied to nextPts and is considered the initial estimate.
159 - **OPTFLOW_LK_GET_MIN_EIGENVALS** use minimum eigen values as an error measure (see
160 minEigThreshold description); if the flag is not set, then L1 distance between patches
161 around the original and a moved point, divided by number of pixels in a window, is used as a
162 error measure.
163@param minEigThreshold the algorithm calculates the minimum eigen value of a 2x2 normal matrix of
164optical flow equations (this matrix is called a spatial gradient matrix in @cite Bouguet00), divided
165by number of pixels in a window; if this value is less than minEigThreshold, then a corresponding
166feature is filtered out and its flow is not processed, so it allows to remove bad points and get a
167performance boost.
168
169The function implements a sparse iterative version of the Lucas-Kanade optical flow in pyramids. See
170@cite Bouguet00 . The function is parallelized with the TBB library.
171
172@note Some examples:
173
174- An example using the Lucas-Kanade optical flow algorithm can be found at
175 opencv_source_code/samples/cpp/lkdemo.cpp
176- (Python) An example using the Lucas-Kanade optical flow algorithm can be found at
177 opencv_source_code/samples/python/lk_track.py
178- (Python) An example using the Lucas-Kanade tracker for homography matching can be found at
179 opencv_source_code/samples/python/lk_homography.py
180 */
181CV_EXPORTS_W void calcOpticalFlowPyrLK( InputArray prevImg, InputArray nextImg,
182 InputArray prevPts, InputOutputArray nextPts,
183 OutputArray status, OutputArray err,
184 Size winSize = Size(21,21), int maxLevel = 3,
185 TermCriteria criteria = TermCriteria(TermCriteria::COUNT+TermCriteria::EPS, 30, 0.01),
186 int flags = 0, double minEigThreshold = 1e-4 );
187
188/** @brief Computes a dense optical flow using the Gunnar Farneback's algorithm.
189
190@param prev first 8-bit single-channel input image.
191@param next second input image of the same size and the same type as prev.
192@param flow computed flow image that has the same size as prev and type CV_32FC2.
193@param pyr_scale parameter, specifying the image scale (\<1) to build pyramids for each image;
194pyr_scale=0.5 means a classical pyramid, where each next layer is twice smaller than the previous
195one.
196@param levels number of pyramid layers including the initial image; levels=1 means that no extra
197layers are created and only the original images are used.
198@param winsize averaging window size; larger values increase the algorithm robustness to image
199noise and give more chances for fast motion detection, but yield more blurred motion field.
200@param iterations number of iterations the algorithm does at each pyramid level.
201@param poly_n size of the pixel neighborhood used to find polynomial expansion in each pixel;
202larger values mean that the image will be approximated with smoother surfaces, yielding more
203robust algorithm and more blurred motion field, typically poly_n =5 or 7.
204@param poly_sigma standard deviation of the Gaussian that is used to smooth derivatives used as a
205basis for the polynomial expansion; for poly_n=5, you can set poly_sigma=1.1, for poly_n=7, a
206good value would be poly_sigma=1.5.
207@param flags operation flags that can be a combination of the following:
208 - **OPTFLOW_USE_INITIAL_FLOW** uses the input flow as an initial flow approximation.
209 - **OPTFLOW_FARNEBACK_GAUSSIAN** uses the Gaussian \f$\texttt{winsize}\times\texttt{winsize}\f$
210 filter instead of a box filter of the same size for optical flow estimation; usually, this
211 option gives z more accurate flow than with a box filter, at the cost of lower speed;
212 normally, winsize for a Gaussian window should be set to a larger value to achieve the same
213 level of robustness.
214
215The function finds an optical flow for each prev pixel using the @cite Farneback2003 algorithm so that
216
217\f[\texttt{prev} (y,x) \sim \texttt{next} ( y + \texttt{flow} (y,x)[1], x + \texttt{flow} (y,x)[0])\f]
218
219@note Some examples:
220
221- An example using the optical flow algorithm described by Gunnar Farneback can be found at
222 opencv_source_code/samples/cpp/fback.cpp
223- (Python) An example using the optical flow algorithm described by Gunnar Farneback can be
224 found at opencv_source_code/samples/python/opt_flow.py
225 */
226CV_EXPORTS_W void calcOpticalFlowFarneback( InputArray prev, InputArray next, InputOutputArray flow,
227 double pyr_scale, int levels, int winsize,
228 int iterations, int poly_n, double poly_sigma,
229 int flags );
230
231/** @brief Computes an optimal affine transformation between two 2D point sets.
232
233@param src First input 2D point set stored in std::vector or Mat, or an image stored in Mat.
234@param dst Second input 2D point set of the same size and the same type as A, or another image.
235@param fullAffine If true, the function finds an optimal affine transformation with no additional
236restrictions (6 degrees of freedom). Otherwise, the class of transformations to choose from is
237limited to combinations of translation, rotation, and uniform scaling (4 degrees of freedom).
238
239The function finds an optimal affine transform *[A|b]* (a 2 x 3 floating-point matrix) that
240approximates best the affine transformation between:
241
242* Two point sets
243* Two raster images. In this case, the function first finds some features in the src image and
244 finds the corresponding features in dst image. After that, the problem is reduced to the first
245 case.
246In case of point sets, the problem is formulated as follows: you need to find a 2x2 matrix *A* and
2472x1 vector *b* so that:
248
249\f[[A^*|b^*] = arg \min _{[A|b]} \sum _i \| \texttt{dst}[i] - A { \texttt{src}[i]}^T - b \| ^2\f]
250where src[i] and dst[i] are the i-th points in src and dst, respectively
251\f$[A|b]\f$ can be either arbitrary (when fullAffine=true ) or have a form of
252\f[\begin{bmatrix} a_{11} & a_{12} & b_1 \\ -a_{12} & a_{11} & b_2 \end{bmatrix}\f]
253when fullAffine=false.
254
255@deprecated Use cv::estimateAffine2D, cv::estimateAffinePartial2D instead. If you are using this function
256with images, extract points using cv::calcOpticalFlowPyrLK and then use the estimation functions.
257
258@sa
259estimateAffine2D, estimateAffinePartial2D, getAffineTransform, getPerspectiveTransform, findHomography
260 */
261CV_DEPRECATED CV_EXPORTS Mat estimateRigidTransform( InputArray src, InputArray dst, bool fullAffine );
262
263enum
264{
265 MOTION_TRANSLATION = 0,
266 MOTION_EUCLIDEAN = 1,
267 MOTION_AFFINE = 2,
268 MOTION_HOMOGRAPHY = 3
269};
270
271/** @brief Computes the Enhanced Correlation Coefficient value between two images @cite EP08 .
272
273@param templateImage single-channel template image; CV_8U or CV_32F array.
274@param inputImage single-channel input image to be warped to provide an image similar to
275 templateImage, same type as templateImage.
276@param inputMask An optional mask to indicate valid values of inputImage.
277
278@sa
279findTransformECC
280 */
281
282CV_EXPORTS_W double computeECC(InputArray templateImage, InputArray inputImage, InputArray inputMask = noArray());
283
284/** @example samples/cpp/image_alignment.cpp
285An example using the image alignment ECC algorithm
286*/
287
288/** @brief Finds the geometric transform (warp) between two images in terms of the ECC criterion @cite EP08 .
289
290@param templateImage single-channel template image; CV_8U or CV_32F array.
291@param inputImage single-channel input image which should be warped with the final warpMatrix in
292order to provide an image similar to templateImage, same type as templateImage.
293@param warpMatrix floating-point \f$2\times 3\f$ or \f$3\times 3\f$ mapping matrix (warp).
294@param motionType parameter, specifying the type of motion:
295 - **MOTION_TRANSLATION** sets a translational motion model; warpMatrix is \f$2\times 3\f$ with
296 the first \f$2\times 2\f$ part being the unity matrix and the rest two parameters being
297 estimated.
298 - **MOTION_EUCLIDEAN** sets a Euclidean (rigid) transformation as motion model; three
299 parameters are estimated; warpMatrix is \f$2\times 3\f$.
300 - **MOTION_AFFINE** sets an affine motion model (DEFAULT); six parameters are estimated;
301 warpMatrix is \f$2\times 3\f$.
302 - **MOTION_HOMOGRAPHY** sets a homography as a motion model; eight parameters are
303 estimated;\`warpMatrix\` is \f$3\times 3\f$.
304@param criteria parameter, specifying the termination criteria of the ECC algorithm;
305criteria.epsilon defines the threshold of the increment in the correlation coefficient between two
306iterations (a negative criteria.epsilon makes criteria.maxcount the only termination criterion).
307Default values are shown in the declaration above.
308@param inputMask An optional mask to indicate valid values of inputImage.
309@param gaussFiltSize An optional value indicating size of gaussian blur filter; (DEFAULT: 5)
310
311The function estimates the optimum transformation (warpMatrix) with respect to ECC criterion
312(@cite EP08), that is
313
314\f[\texttt{warpMatrix} = \arg\max_{W} \texttt{ECC}(\texttt{templateImage}(x,y),\texttt{inputImage}(x',y'))\f]
315
316where
317
318\f[\begin{bmatrix} x' \\ y' \end{bmatrix} = W \cdot \begin{bmatrix} x \\ y \\ 1 \end{bmatrix}\f]
319
320(the equation holds with homogeneous coordinates for homography). It returns the final enhanced
321correlation coefficient, that is the correlation coefficient between the template image and the
322final warped input image. When a \f$3\times 3\f$ matrix is given with motionType =0, 1 or 2, the third
323row is ignored.
324
325Unlike findHomography and estimateRigidTransform, the function findTransformECC implements an
326area-based alignment that builds on intensity similarities. In essence, the function updates the
327initial transformation that roughly aligns the images. If this information is missing, the identity
328warp (unity matrix) is used as an initialization. Note that if images undergo strong
329displacements/rotations, an initial transformation that roughly aligns the images is necessary
330(e.g., a simple euclidean/similarity transform that allows for the images showing the same image
331content approximately). Use inverse warping in the second image to take an image close to the first
332one, i.e. use the flag WARP_INVERSE_MAP with warpAffine or warpPerspective. See also the OpenCV
333sample image_alignment.cpp that demonstrates the use of the function. Note that the function throws
334an exception if algorithm does not converges.
335
336@sa
337computeECC, estimateAffine2D, estimateAffinePartial2D, findHomography
338 */
339CV_EXPORTS_W double findTransformECC( InputArray templateImage, InputArray inputImage,
340 InputOutputArray warpMatrix, int motionType,
341 TermCriteria criteria,
342 InputArray inputMask, int gaussFiltSize);
343
344/** @overload */
345CV_EXPORTS_W
346double findTransformECC(InputArray templateImage, InputArray inputImage,
347 InputOutputArray warpMatrix, int motionType = MOTION_AFFINE,
348 TermCriteria criteria = TermCriteria(TermCriteria::COUNT+TermCriteria::EPS, 50, 0.001),
349 InputArray inputMask = noArray());
350
351/** @example samples/cpp/kalman.cpp
352An example using the standard Kalman filter
353*/
354
355/** @brief Kalman filter class.
356
357The class implements a standard Kalman filter <http://en.wikipedia.org/wiki/Kalman_filter>,
358@cite Welch95 . However, you can modify transitionMatrix, controlMatrix, and measurementMatrix to get
359an extended Kalman filter functionality.
360@note In C API when CvKalman\* kalmanFilter structure is not needed anymore, it should be released
361with cvReleaseKalman(&kalmanFilter)
362 */
363class CV_EXPORTS_W KalmanFilter
364{
365public:
366 CV_WRAP KalmanFilter();
367 /** @overload
368 @param dynamParams Dimensionality of the state.
369 @param measureParams Dimensionality of the measurement.
370 @param controlParams Dimensionality of the control vector.
371 @param type Type of the created matrices that should be CV_32F or CV_64F.
372 */
373 CV_WRAP KalmanFilter( int dynamParams, int measureParams, int controlParams = 0, int type = CV_32F );
374
375 /** @brief Re-initializes Kalman filter. The previous content is destroyed.
376
377 @param dynamParams Dimensionality of the state.
378 @param measureParams Dimensionality of the measurement.
379 @param controlParams Dimensionality of the control vector.
380 @param type Type of the created matrices that should be CV_32F or CV_64F.
381 */
382 void init( int dynamParams, int measureParams, int controlParams = 0, int type = CV_32F );
383
384 /** @brief Computes a predicted state.
385
386 @param control The optional input control
387 */
388 CV_WRAP const Mat& predict( const Mat& control = Mat() );
389
390 /** @brief Updates the predicted state from the measurement.
391
392 @param measurement The measured system parameters
393 */
394 CV_WRAP const Mat& correct( const Mat& measurement );
395
396 CV_PROP_RW Mat statePre; //!< predicted state (x'(k)): x(k)=A*x(k-1)+B*u(k)
397 CV_PROP_RW Mat statePost; //!< corrected state (x(k)): x(k)=x'(k)+K(k)*(z(k)-H*x'(k))
398 CV_PROP_RW Mat transitionMatrix; //!< state transition matrix (A)
399 CV_PROP_RW Mat controlMatrix; //!< control matrix (B) (not used if there is no control)
400 CV_PROP_RW Mat measurementMatrix; //!< measurement matrix (H)
401 CV_PROP_RW Mat processNoiseCov; //!< process noise covariance matrix (Q)
402 CV_PROP_RW Mat measurementNoiseCov;//!< measurement noise covariance matrix (R)
403 CV_PROP_RW Mat errorCovPre; //!< priori error estimate covariance matrix (P'(k)): P'(k)=A*P(k-1)*At + Q)*/
404 CV_PROP_RW Mat gain; //!< Kalman gain matrix (K(k)): K(k)=P'(k)*Ht*inv(H*P'(k)*Ht+R)
405 CV_PROP_RW Mat errorCovPost; //!< posteriori error estimate covariance matrix (P(k)): P(k)=(I-K(k)*H)*P'(k)
406
407 // temporary matrices
408 Mat temp1;
409 Mat temp2;
410 Mat temp3;
411 Mat temp4;
412 Mat temp5;
413};
414
415
416/** @brief Read a .flo file
417
418 @param path Path to the file to be loaded
419
420 The function readOpticalFlow loads a flow field from a file and returns it as a single matrix.
421 Resulting Mat has a type CV_32FC2 - floating-point, 2-channel. First channel corresponds to the
422 flow in the horizontal direction (u), second - vertical (v).
423 */
424CV_EXPORTS_W Mat readOpticalFlow( const String& path );
425/** @brief Write a .flo to disk
426
427 @param path Path to the file to be written
428 @param flow Flow field to be stored
429
430 The function stores a flow field in a file, returns true on success, false otherwise.
431 The flow field must be a 2-channel, floating-point matrix (CV_32FC2). First channel corresponds
432 to the flow in the horizontal direction (u), second - vertical (v).
433 */
434CV_EXPORTS_W bool writeOpticalFlow( const String& path, InputArray flow );
435
436/**
437 Base class for dense optical flow algorithms
438*/
439class CV_EXPORTS_W DenseOpticalFlow : public Algorithm
440{
441public:
442 /** @brief Calculates an optical flow.
443
444 @param I0 first 8-bit single-channel input image.
445 @param I1 second input image of the same size and the same type as prev.
446 @param flow computed flow image that has the same size as prev and type CV_32FC2.
447 */
448 CV_WRAP virtual void calc( InputArray I0, InputArray I1, InputOutputArray flow ) = 0;
449 /** @brief Releases all inner buffers.
450 */
451 CV_WRAP virtual void collectGarbage() = 0;
452};
453
454/** @brief Base interface for sparse optical flow algorithms.
455 */
456class CV_EXPORTS_W SparseOpticalFlow : public Algorithm
457{
458public:
459 /** @brief Calculates a sparse optical flow.
460
461 @param prevImg First input image.
462 @param nextImg Second input image of the same size and the same type as prevImg.
463 @param prevPts Vector of 2D points for which the flow needs to be found.
464 @param nextPts Output vector of 2D points containing the calculated new positions of input features in the second image.
465 @param status Output status vector. Each element of the vector is set to 1 if the
466 flow for the corresponding features has been found. Otherwise, it is set to 0.
467 @param err Optional output vector that contains error response for each point (inverse confidence).
468 */
469 CV_WRAP virtual void calc(InputArray prevImg, InputArray nextImg,
470 InputArray prevPts, InputOutputArray nextPts,
471 OutputArray status,
472 OutputArray err = cv::noArray()) = 0;
473};
474
475
476/** @brief Class computing a dense optical flow using the Gunnar Farneback's algorithm.
477 */
478class CV_EXPORTS_W FarnebackOpticalFlow : public DenseOpticalFlow
479{
480public:
481 CV_WRAP virtual int getNumLevels() const = 0;
482 CV_WRAP virtual void setNumLevels(int numLevels) = 0;
483
484 CV_WRAP virtual double getPyrScale() const = 0;
485 CV_WRAP virtual void setPyrScale(double pyrScale) = 0;
486
487 CV_WRAP virtual bool getFastPyramids() const = 0;
488 CV_WRAP virtual void setFastPyramids(bool fastPyramids) = 0;
489
490 CV_WRAP virtual int getWinSize() const = 0;
491 CV_WRAP virtual void setWinSize(int winSize) = 0;
492
493 CV_WRAP virtual int getNumIters() const = 0;
494 CV_WRAP virtual void setNumIters(int numIters) = 0;
495
496 CV_WRAP virtual int getPolyN() const = 0;
497 CV_WRAP virtual void setPolyN(int polyN) = 0;
498
499 CV_WRAP virtual double getPolySigma() const = 0;
500 CV_WRAP virtual void setPolySigma(double polySigma) = 0;
501
502 CV_WRAP virtual int getFlags() const = 0;
503 CV_WRAP virtual void setFlags(int flags) = 0;
504
505 CV_WRAP static Ptr<FarnebackOpticalFlow> create(
506 int numLevels = 5,
507 double pyrScale = 0.5,
508 bool fastPyramids = false,
509 int winSize = 13,
510 int numIters = 10,
511 int polyN = 5,
512 double polySigma = 1.1,
513 int flags = 0);
514};
515
516/** @brief Variational optical flow refinement
517
518This class implements variational refinement of the input flow field, i.e.
519it uses input flow to initialize the minimization of the following functional:
520\f$E(U) = \int_{\Omega} \delta \Psi(E_I) + \gamma \Psi(E_G) + \alpha \Psi(E_S) \f$,
521where \f$E_I,E_G,E_S\f$ are color constancy, gradient constancy and smoothness terms
522respectively. \f$\Psi(s^2)=\sqrt{s^2+\epsilon^2}\f$ is a robust penalizer to limit the
523influence of outliers. A complete formulation and a description of the minimization
524procedure can be found in @cite Brox2004
525*/
526class CV_EXPORTS_W VariationalRefinement : public DenseOpticalFlow
527{
528public:
529 /** @brief @ref calc function overload to handle separate horizontal (u) and vertical (v) flow components
530 (to avoid extra splits/merges) */
531 CV_WRAP virtual void calcUV(InputArray I0, InputArray I1, InputOutputArray flow_u, InputOutputArray flow_v) = 0;
532
533 /** @brief Number of outer (fixed-point) iterations in the minimization procedure.
534 @see setFixedPointIterations */
535 CV_WRAP virtual int getFixedPointIterations() const = 0;
536 /** @copybrief getFixedPointIterations @see getFixedPointIterations */
537 CV_WRAP virtual void setFixedPointIterations(int val) = 0;
538
539 /** @brief Number of inner successive over-relaxation (SOR) iterations
540 in the minimization procedure to solve the respective linear system.
541 @see setSorIterations */
542 CV_WRAP virtual int getSorIterations() const = 0;
543 /** @copybrief getSorIterations @see getSorIterations */
544 CV_WRAP virtual void setSorIterations(int val) = 0;
545
546 /** @brief Relaxation factor in SOR
547 @see setOmega */
548 CV_WRAP virtual float getOmega() const = 0;
549 /** @copybrief getOmega @see getOmega */
550 CV_WRAP virtual void setOmega(float val) = 0;
551
552 /** @brief Weight of the smoothness term
553 @see setAlpha */
554 CV_WRAP virtual float getAlpha() const = 0;
555 /** @copybrief getAlpha @see getAlpha */
556 CV_WRAP virtual void setAlpha(float val) = 0;
557
558 /** @brief Weight of the color constancy term
559 @see setDelta */
560 CV_WRAP virtual float getDelta() const = 0;
561 /** @copybrief getDelta @see getDelta */
562 CV_WRAP virtual void setDelta(float val) = 0;
563
564 /** @brief Weight of the gradient constancy term
565 @see setGamma */
566 CV_WRAP virtual float getGamma() const = 0;
567 /** @copybrief getGamma @see getGamma */
568 CV_WRAP virtual void setGamma(float val) = 0;
569
570 /** @brief Norm value shift for robust penalizer
571 @see setEpsilon */
572 CV_WRAP virtual float getEpsilon() const = 0;
573 /** @copybrief getEpsilon @see getEpsilon */
574 CV_WRAP virtual void setEpsilon(float val) = 0;
575
576 /** @brief Creates an instance of VariationalRefinement
577 */
578 CV_WRAP static Ptr<VariationalRefinement> create();
579};
580
581/** @brief DIS optical flow algorithm.
582
583This class implements the Dense Inverse Search (DIS) optical flow algorithm. More
584details about the algorithm can be found at @cite Kroeger2016 . Includes three presets with preselected
585parameters to provide reasonable trade-off between speed and quality. However, even the slowest preset is
586still relatively fast, use DeepFlow if you need better quality and don't care about speed.
587
588This implementation includes several additional features compared to the algorithm described in the paper,
589including spatial propagation of flow vectors (@ref getUseSpatialPropagation), as well as an option to
590utilize an initial flow approximation passed to @ref calc (which is, essentially, temporal propagation,
591if the previous frame's flow field is passed).
592*/
593class CV_EXPORTS_W DISOpticalFlow : public DenseOpticalFlow
594{
595public:
596 enum
597 {
598 PRESET_ULTRAFAST = 0,
599 PRESET_FAST = 1,
600 PRESET_MEDIUM = 2
601 };
602
603 /** @brief Finest level of the Gaussian pyramid on which the flow is computed (zero level
604 corresponds to the original image resolution). The final flow is obtained by bilinear upscaling.
605 @see setFinestScale */
606 CV_WRAP virtual int getFinestScale() const = 0;
607 /** @copybrief getFinestScale @see getFinestScale */
608 CV_WRAP virtual void setFinestScale(int val) = 0;
609
610 /** @brief Size of an image patch for matching (in pixels). Normally, default 8x8 patches work well
611 enough in most cases.
612 @see setPatchSize */
613 CV_WRAP virtual int getPatchSize() const = 0;
614 /** @copybrief getPatchSize @see getPatchSize */
615 CV_WRAP virtual void setPatchSize(int val) = 0;
616
617 /** @brief Stride between neighbor patches. Must be less than patch size. Lower values correspond
618 to higher flow quality.
619 @see setPatchStride */
620 CV_WRAP virtual int getPatchStride() const = 0;
621 /** @copybrief getPatchStride @see getPatchStride */
622 CV_WRAP virtual void setPatchStride(int val) = 0;
623
624 /** @brief Maximum number of gradient descent iterations in the patch inverse search stage. Higher values
625 may improve quality in some cases.
626 @see setGradientDescentIterations */
627 CV_WRAP virtual int getGradientDescentIterations() const = 0;
628 /** @copybrief getGradientDescentIterations @see getGradientDescentIterations */
629 CV_WRAP virtual void setGradientDescentIterations(int val) = 0;
630
631 /** @brief Number of fixed point iterations of variational refinement per scale. Set to zero to
632 disable variational refinement completely. Higher values will typically result in more smooth and
633 high-quality flow.
634 @see setGradientDescentIterations */
635 CV_WRAP virtual int getVariationalRefinementIterations() const = 0;
636 /** @copybrief getGradientDescentIterations @see getGradientDescentIterations */
637 CV_WRAP virtual void setVariationalRefinementIterations(int val) = 0;
638
639 /** @brief Weight of the smoothness term
640 @see setVariationalRefinementAlpha */
641 CV_WRAP virtual float getVariationalRefinementAlpha() const = 0;
642 /** @copybrief getVariationalRefinementAlpha @see getVariationalRefinementAlpha */
643 CV_WRAP virtual void setVariationalRefinementAlpha(float val) = 0;
644
645 /** @brief Weight of the color constancy term
646 @see setVariationalRefinementDelta */
647 CV_WRAP virtual float getVariationalRefinementDelta() const = 0;
648 /** @copybrief getVariationalRefinementDelta @see getVariationalRefinementDelta */
649 CV_WRAP virtual void setVariationalRefinementDelta(float val) = 0;
650
651 /** @brief Weight of the gradient constancy term
652 @see setVariationalRefinementGamma */
653 CV_WRAP virtual float getVariationalRefinementGamma() const = 0;
654 /** @copybrief getVariationalRefinementGamma @see getVariationalRefinementGamma */
655 CV_WRAP virtual void setVariationalRefinementGamma(float val) = 0;
656
657 /** @brief Norm value shift for robust penalizer
658 @see setVariationalRefinementEpsilon */
659 CV_WRAP virtual float getVariationalRefinementEpsilon() const = 0;
660 /** @copybrief getVariationalRefinementEpsilon @see getVariationalRefinementEpsilon */
661 CV_WRAP virtual void setVariationalRefinementEpsilon(float val) = 0;
662
663
664 /** @brief Whether to use mean-normalization of patches when computing patch distance. It is turned on
665 by default as it typically provides a noticeable quality boost because of increased robustness to
666 illumination variations. Turn it off if you are certain that your sequence doesn't contain any changes
667 in illumination.
668 @see setUseMeanNormalization */
669 CV_WRAP virtual bool getUseMeanNormalization() const = 0;
670 /** @copybrief getUseMeanNormalization @see getUseMeanNormalization */
671 CV_WRAP virtual void setUseMeanNormalization(bool val) = 0;
672
673 /** @brief Whether to use spatial propagation of good optical flow vectors. This option is turned on by
674 default, as it tends to work better on average and can sometimes help recover from major errors
675 introduced by the coarse-to-fine scheme employed by the DIS optical flow algorithm. Turning this
676 option off can make the output flow field a bit smoother, however.
677 @see setUseSpatialPropagation */
678 CV_WRAP virtual bool getUseSpatialPropagation() const = 0;
679 /** @copybrief getUseSpatialPropagation @see getUseSpatialPropagation */
680 CV_WRAP virtual void setUseSpatialPropagation(bool val) = 0;
681
682 /** @brief Creates an instance of DISOpticalFlow
683
684 @param preset one of PRESET_ULTRAFAST, PRESET_FAST and PRESET_MEDIUM
685 */
686 CV_WRAP static Ptr<DISOpticalFlow> create(int preset = DISOpticalFlow::PRESET_FAST);
687};
688
689/** @brief Class used for calculating a sparse optical flow.
690
691The class can calculate an optical flow for a sparse feature set using the
692iterative Lucas-Kanade method with pyramids.
693
694@sa calcOpticalFlowPyrLK
695
696*/
697class CV_EXPORTS_W SparsePyrLKOpticalFlow : public SparseOpticalFlow
698{
699public:
700 CV_WRAP virtual Size getWinSize() const = 0;
701 CV_WRAP virtual void setWinSize(Size winSize) = 0;
702
703 CV_WRAP virtual int getMaxLevel() const = 0;
704 CV_WRAP virtual void setMaxLevel(int maxLevel) = 0;
705
706 CV_WRAP virtual TermCriteria getTermCriteria() const = 0;
707 CV_WRAP virtual void setTermCriteria(TermCriteria& crit) = 0;
708
709 CV_WRAP virtual int getFlags() const = 0;
710 CV_WRAP virtual void setFlags(int flags) = 0;
711
712 CV_WRAP virtual double getMinEigThreshold() const = 0;
713 CV_WRAP virtual void setMinEigThreshold(double minEigThreshold) = 0;
714
715 CV_WRAP static Ptr<SparsePyrLKOpticalFlow> create(
716 Size winSize = Size(21, 21),
717 int maxLevel = 3, TermCriteria crit =
718 TermCriteria(TermCriteria::COUNT+TermCriteria::EPS, 30, 0.01),
719 int flags = 0,
720 double minEigThreshold = 1e-4);
721};
722
723
724
725
726/** @brief Base abstract class for the long-term tracker
727 */
728class CV_EXPORTS_W Tracker
729{
730protected:
731 Tracker();
732public:
733 virtual ~Tracker();
734
735 /** @brief Initialize the tracker with a known bounding box that surrounded the target
736 @param image The initial frame
737 @param boundingBox The initial bounding box
738 */
739 CV_WRAP virtual
740 void init(InputArray image, const Rect& boundingBox) = 0;
741
742 /** @brief Update the tracker, find the new most likely bounding box for the target
743 @param image The current frame
744 @param boundingBox The bounding box that represent the new target location, if true was returned, not
745 modified otherwise
746
747 @return True means that target was located and false means that tracker cannot locate target in
748 current frame. Note, that latter *does not* imply that tracker has failed, maybe target is indeed
749 missing from the frame (say, out of sight)
750 */
751 CV_WRAP virtual
752 bool update(InputArray image, CV_OUT Rect& boundingBox) = 0;
753};
754
755
756
757/** @brief The MIL algorithm trains a classifier in an online manner to separate the object from the
758background.
759
760Multiple Instance Learning avoids the drift problem for a robust tracking. The implementation is
761based on @cite MIL .
762
763Original code can be found here <http://vision.ucsd.edu/~bbabenko/project_miltrack.shtml>
764 */
765class CV_EXPORTS_W TrackerMIL : public Tracker
766{
767protected:
768 TrackerMIL(); // use ::create()
769public:
770 virtual ~TrackerMIL() CV_OVERRIDE;
771
772 struct CV_EXPORTS_W_SIMPLE Params
773 {
774 CV_WRAP Params();
775 //parameters for sampler
776 CV_PROP_RW float samplerInitInRadius; //!< radius for gathering positive instances during init
777 CV_PROP_RW int samplerInitMaxNegNum; //!< # negative samples to use during init
778 CV_PROP_RW float samplerSearchWinSize; //!< size of search window
779 CV_PROP_RW float samplerTrackInRadius; //!< radius for gathering positive instances during tracking
780 CV_PROP_RW int samplerTrackMaxPosNum; //!< # positive samples to use during tracking
781 CV_PROP_RW int samplerTrackMaxNegNum; //!< # negative samples to use during tracking
782 CV_PROP_RW int featureSetNumFeatures; //!< # features
783 };
784
785 /** @brief Create MIL tracker instance
786 * @param parameters MIL parameters TrackerMIL::Params
787 */
788 static CV_WRAP
789 Ptr<TrackerMIL> create(const TrackerMIL::Params &parameters = TrackerMIL::Params());
790
791 //void init(InputArray image, const Rect& boundingBox) CV_OVERRIDE;
792 //bool update(InputArray image, CV_OUT Rect& boundingBox) CV_OVERRIDE;
793};
794
795
796
797/** @brief the GOTURN (Generic Object Tracking Using Regression Networks) tracker
798 *
799 * GOTURN (@cite GOTURN) is kind of trackers based on Convolutional Neural Networks (CNN). While taking all advantages of CNN trackers,
800 * GOTURN is much faster due to offline training without online fine-tuning nature.
801 * GOTURN tracker addresses the problem of single target tracking: given a bounding box label of an object in the first frame of the video,
802 * we track that object through the rest of the video. NOTE: Current method of GOTURN does not handle occlusions; however, it is fairly
803 * robust to viewpoint changes, lighting changes, and deformations.
804 * Inputs of GOTURN are two RGB patches representing Target and Search patches resized to 227x227.
805 * Outputs of GOTURN are predicted bounding box coordinates, relative to Search patch coordinate system, in format X1,Y1,X2,Y2.
806 * Original paper is here: <http://davheld.github.io/GOTURN/GOTURN.pdf>
807 * As long as original authors implementation: <https://github.com/davheld/GOTURN#train-the-tracker>
808 * Implementation of training algorithm is placed in separately here due to 3d-party dependencies:
809 * <https://github.com/Auron-X/GOTURN_Training_Toolkit>
810 * GOTURN architecture goturn.prototxt and trained model goturn.caffemodel are accessible on opencv_extra GitHub repository.
811 */
812class CV_EXPORTS_W TrackerGOTURN : public Tracker
813{
814protected:
815 TrackerGOTURN(); // use ::create()
816public:
817 virtual ~TrackerGOTURN() CV_OVERRIDE;
818
819 struct CV_EXPORTS_W_SIMPLE Params
820 {
821 CV_WRAP Params();
822 CV_PROP_RW std::string modelTxt;
823 CV_PROP_RW std::string modelBin;
824 };
825
826 /** @brief Constructor
827 @param parameters GOTURN parameters TrackerGOTURN::Params
828 */
829 static CV_WRAP
830 Ptr<TrackerGOTURN> create(const TrackerGOTURN::Params& parameters = TrackerGOTURN::Params());
831
832#ifdef HAVE_OPENCV_DNN
833 /** @brief Constructor
834 @param model pre-loaded GOTURN model
835 */
836 static CV_WRAP Ptr<TrackerGOTURN> create(const dnn::Net& model);
837#endif
838
839 //void init(InputArray image, const Rect& boundingBox) CV_OVERRIDE;
840 //bool update(InputArray image, CV_OUT Rect& boundingBox) CV_OVERRIDE;
841};
842
843class CV_EXPORTS_W TrackerDaSiamRPN : public Tracker
844{
845protected:
846 TrackerDaSiamRPN(); // use ::create()
847public:
848 virtual ~TrackerDaSiamRPN() CV_OVERRIDE;
849
850 struct CV_EXPORTS_W_SIMPLE Params
851 {
852 CV_WRAP Params();
853 CV_PROP_RW std::string model;
854 CV_PROP_RW std::string kernel_cls1;
855 CV_PROP_RW std::string kernel_r1;
856 CV_PROP_RW int backend;
857 CV_PROP_RW int target;
858 };
859
860 /** @brief Constructor
861 @param parameters DaSiamRPN parameters TrackerDaSiamRPN::Params
862 */
863 static CV_WRAP
864 Ptr<TrackerDaSiamRPN> create(const TrackerDaSiamRPN::Params& parameters = TrackerDaSiamRPN::Params());
865
866#ifdef HAVE_OPENCV_DNN
867 /** @brief Constructor
868 * @param siam_rpn pre-loaded SiamRPN model
869 * @param kernel_cls1 pre-loaded CLS model
870 * @param kernel_r1 pre-loaded R1 model
871 */
872 static CV_WRAP
873 Ptr<TrackerDaSiamRPN> create(const dnn::Net& siam_rpn, const dnn::Net& kernel_cls1, const dnn::Net& kernel_r1);
874#endif
875
876 /** @brief Return tracking score
877 */
878 CV_WRAP virtual float getTrackingScore() = 0;
879
880 //void init(InputArray image, const Rect& boundingBox) CV_OVERRIDE;
881 //bool update(InputArray image, CV_OUT Rect& boundingBox) CV_OVERRIDE;
882};
883
884/** @brief the Nano tracker is a super lightweight dnn-based general object tracking.
885 *
886 * Nano tracker is much faster and extremely lightweight due to special model structure, the whole model size is about 1.9 MB.
887 * Nano tracker needs two models: one for feature extraction (backbone) and the another for localization (neckhead).
888 * Model download link: https://github.com/HonglinChu/SiamTrackers/tree/master/NanoTrack/models/nanotrackv2
889 * Original repo is here: https://github.com/HonglinChu/NanoTrack
890 * Author: HongLinChu, 1628464345@qq.com
891 */
892class CV_EXPORTS_W TrackerNano : public Tracker
893{
894protected:
895 TrackerNano(); // use ::create()
896public:
897 virtual ~TrackerNano() CV_OVERRIDE;
898
899 struct CV_EXPORTS_W_SIMPLE Params
900 {
901 CV_WRAP Params();
902 CV_PROP_RW std::string backbone;
903 CV_PROP_RW std::string neckhead;
904 CV_PROP_RW int backend;
905 CV_PROP_RW int target;
906 };
907
908 /** @brief Constructor
909 @param parameters NanoTrack parameters TrackerNano::Params
910 */
911 static CV_WRAP
912 Ptr<TrackerNano> create(const TrackerNano::Params& parameters = TrackerNano::Params());
913
914#ifdef HAVE_OPENCV_DNN
915 /** @brief Constructor
916 * @param backbone pre-loaded backbone model
917 * @param neckhead pre-loaded neckhead model
918 */
919 static CV_WRAP
920 Ptr<TrackerNano> create(const dnn::Net& backbone, const dnn::Net& neckhead);
921#endif
922
923 /** @brief Return tracking score
924 */
925 CV_WRAP virtual float getTrackingScore() = 0;
926
927 //void init(InputArray image, const Rect& boundingBox) CV_OVERRIDE;
928 //bool update(InputArray image, CV_OUT Rect& boundingBox) CV_OVERRIDE;
929};
930
931/** @brief the VIT tracker is a super lightweight dnn-based general object tracking.
932 *
933 * VIT tracker is much faster and extremely lightweight due to special model structure, the model file is about 767KB.
934 * Model download link: https://github.com/opencv/opencv_zoo/tree/main/models/object_tracking_vittrack
935 * Author: PengyuLiu, 1872918507@qq.com
936 */
937class CV_EXPORTS_W TrackerVit : public Tracker
938{
939protected:
940 TrackerVit(); // use ::create()
941public:
942 virtual ~TrackerVit() CV_OVERRIDE;
943
944 struct CV_EXPORTS_W_SIMPLE Params
945 {
946 CV_WRAP Params();
947 CV_PROP_RW std::string net;
948 CV_PROP_RW int backend;
949 CV_PROP_RW int target;
950 CV_PROP_RW Scalar meanvalue;
951 CV_PROP_RW Scalar stdvalue;
952 CV_PROP_RW float tracking_score_threshold;
953 };
954
955 /** @brief Constructor
956 @param parameters vit tracker parameters TrackerVit::Params
957 */
958 static CV_WRAP
959 Ptr<TrackerVit> create(const TrackerVit::Params& parameters = TrackerVit::Params());
960
961#ifdef HAVE_OPENCV_DNN
962 /** @brief Constructor
963 * @param model pre-loaded DNN model
964 * @param meanvalue mean value for image preprocessing
965 * @param stdvalue std value for image preprocessing
966 * @param tracking_score_threshold threshold for tracking score
967 */
968 static CV_WRAP
969 Ptr<TrackerVit> create(const dnn::Net& model, Scalar meanvalue = Scalar(0.485, 0.456, 0.406),
970 Scalar stdvalue = Scalar(0.229, 0.224, 0.225), float tracking_score_threshold = 0.20f);
971#endif
972
973 /** @brief Return tracking score
974 */
975 CV_WRAP virtual float getTrackingScore() = 0;
976
977 // void init(InputArray image, const Rect& boundingBox) CV_OVERRIDE;
978 // bool update(InputArray image, CV_OUT Rect& boundingBox) CV_OVERRIDE;
979};
980
981//! @} video_track
982
983} // cv
984
985#endif
986

source code of opencv/modules/video/include/opencv2/video/tracking.hpp