• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 // 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 
45 #if !defined HAVE_CUDA || defined(CUDA_DISABLER)
46 
create(double,double,double,int,int,double,int,double,double,bool)47 Ptr<OpticalFlowDual_TVL1> cv::cuda::OpticalFlowDual_TVL1::create(double, double, double, int, int, double, int, double, double, bool) { throw_no_cuda(); return Ptr<OpticalFlowDual_TVL1>(); }
48 
49 #else
50 
51 using namespace cv;
52 using namespace cv::cuda;
53 
54 namespace tvl1flow
55 {
56     void centeredGradient(PtrStepSzf src, PtrStepSzf dx, PtrStepSzf dy, cudaStream_t stream);
57     void warpBackward(PtrStepSzf I0, PtrStepSzf I1, PtrStepSzf I1x, PtrStepSzf I1y,
58                       PtrStepSzf u1, PtrStepSzf u2,
59                       PtrStepSzf I1w, PtrStepSzf I1wx, PtrStepSzf I1wy,
60                       PtrStepSzf grad, PtrStepSzf rho,
61                       cudaStream_t stream);
62     void estimateU(PtrStepSzf I1wx, PtrStepSzf I1wy,
63                    PtrStepSzf grad, PtrStepSzf rho_c,
64                    PtrStepSzf p11, PtrStepSzf p12, PtrStepSzf p21, PtrStepSzf p22, PtrStepSzf p31, PtrStepSzf p32,
65                    PtrStepSzf u1, PtrStepSzf u2, PtrStepSzf u3, PtrStepSzf error,
66                    float l_t, float theta, float gamma, bool calcError,
67                    cudaStream_t stream);
68     void estimateDualVariables(PtrStepSzf u1, PtrStepSzf u2, PtrStepSzf u3,
69                                PtrStepSzf p11, PtrStepSzf p12, PtrStepSzf p21, PtrStepSzf p22, PtrStepSzf p31, PtrStepSzf p32,
70                                float taut, float gamma,
71                                cudaStream_t stream);
72 }
73 
74 namespace
75 {
76     class OpticalFlowDual_TVL1_Impl : public OpticalFlowDual_TVL1
77     {
78     public:
OpticalFlowDual_TVL1_Impl(double tau,double lambda,double theta,int nscales,int warps,double epsilon,int iterations,double scaleStep,double gamma,bool useInitialFlow)79         OpticalFlowDual_TVL1_Impl(double tau, double lambda, double theta, int nscales, int warps, double epsilon,
80                                   int iterations, double scaleStep, double gamma, bool useInitialFlow) :
81             tau_(tau), lambda_(lambda), gamma_(gamma), theta_(theta), nscales_(nscales), warps_(warps),
82             epsilon_(epsilon), iterations_(iterations), scaleStep_(scaleStep), useInitialFlow_(useInitialFlow)
83         {
84         }
85 
getTau() const86         virtual double getTau() const { return tau_; }
setTau(double tau)87         virtual void setTau(double tau) { tau_ = tau; }
88 
getLambda() const89         virtual double getLambda() const { return lambda_; }
setLambda(double lambda)90         virtual void setLambda(double lambda) { lambda_ = lambda; }
91 
getGamma() const92         virtual double getGamma() const { return gamma_; }
setGamma(double gamma)93         virtual void setGamma(double gamma) { gamma_ = gamma; }
94 
getTheta() const95         virtual double getTheta() const { return theta_; }
setTheta(double theta)96         virtual void setTheta(double theta) { theta_ = theta; }
97 
getNumScales() const98         virtual int getNumScales() const { return nscales_; }
setNumScales(int nscales)99         virtual void setNumScales(int nscales) { nscales_ = nscales; }
100 
getNumWarps() const101         virtual int getNumWarps() const { return warps_; }
setNumWarps(int warps)102         virtual void setNumWarps(int warps) { warps_ = warps; }
103 
getEpsilon() const104         virtual double getEpsilon() const { return epsilon_; }
setEpsilon(double epsilon)105         virtual void setEpsilon(double epsilon) { epsilon_ = epsilon; }
106 
getNumIterations() const107         virtual int getNumIterations() const { return iterations_; }
setNumIterations(int iterations)108         virtual void setNumIterations(int iterations) { iterations_ = iterations; }
109 
getScaleStep() const110         virtual double getScaleStep() const { return scaleStep_; }
setScaleStep(double scaleStep)111         virtual void setScaleStep(double scaleStep) { scaleStep_ = scaleStep; }
112 
getUseInitialFlow() const113         virtual bool getUseInitialFlow() const { return useInitialFlow_; }
setUseInitialFlow(bool useInitialFlow)114         virtual void setUseInitialFlow(bool useInitialFlow) { useInitialFlow_ = useInitialFlow; }
115 
116         virtual void calc(InputArray I0, InputArray I1, InputOutputArray flow, Stream& stream);
117 
118     private:
119         double tau_;
120         double lambda_;
121         double gamma_;
122         double theta_;
123         int nscales_;
124         int warps_;
125         double epsilon_;
126         int iterations_;
127         double scaleStep_;
128         bool useInitialFlow_;
129 
130     private:
131         void calcImpl(const GpuMat& I0, const GpuMat& I1, GpuMat& flowx, GpuMat& flowy, Stream& stream);
132         void procOneScale(const GpuMat& I0, const GpuMat& I1, GpuMat& u1, GpuMat& u2, GpuMat& u3, Stream& stream);
133 
134         std::vector<GpuMat> I0s;
135         std::vector<GpuMat> I1s;
136         std::vector<GpuMat> u1s;
137         std::vector<GpuMat> u2s;
138         std::vector<GpuMat> u3s;
139 
140         GpuMat I1x_buf;
141         GpuMat I1y_buf;
142 
143         GpuMat I1w_buf;
144         GpuMat I1wx_buf;
145         GpuMat I1wy_buf;
146 
147         GpuMat grad_buf;
148         GpuMat rho_c_buf;
149 
150         GpuMat p11_buf;
151         GpuMat p12_buf;
152         GpuMat p21_buf;
153         GpuMat p22_buf;
154         GpuMat p31_buf;
155         GpuMat p32_buf;
156 
157         GpuMat diff_buf;
158         GpuMat norm_buf;
159     };
160 
calc(InputArray _frame0,InputArray _frame1,InputOutputArray _flow,Stream & stream)161     void OpticalFlowDual_TVL1_Impl::calc(InputArray _frame0, InputArray _frame1, InputOutputArray _flow, Stream& stream)
162     {
163         const GpuMat frame0 = _frame0.getGpuMat();
164         const GpuMat frame1 = _frame1.getGpuMat();
165 
166         BufferPool pool(stream);
167         GpuMat flowx = pool.getBuffer(frame0.size(), CV_32FC1);
168         GpuMat flowy = pool.getBuffer(frame0.size(), CV_32FC1);
169 
170         calcImpl(frame0, frame1, flowx, flowy, stream);
171 
172         GpuMat flows[] = {flowx, flowy};
173         cuda::merge(flows, 2, _flow, stream);
174     }
175 
calcImpl(const GpuMat & I0,const GpuMat & I1,GpuMat & flowx,GpuMat & flowy,Stream & stream)176     void OpticalFlowDual_TVL1_Impl::calcImpl(const GpuMat& I0, const GpuMat& I1, GpuMat& flowx, GpuMat& flowy, Stream& stream)
177     {
178         CV_Assert( I0.type() == CV_8UC1 || I0.type() == CV_32FC1 );
179         CV_Assert( I0.size() == I1.size() );
180         CV_Assert( I0.type() == I1.type() );
181         CV_Assert( !useInitialFlow_ || (flowx.size() == I0.size() && flowx.type() == CV_32FC1 && flowy.size() == flowx.size() && flowy.type() == flowx.type()) );
182         CV_Assert( nscales_ > 0 );
183 
184         // allocate memory for the pyramid structure
185         I0s.resize(nscales_);
186         I1s.resize(nscales_);
187         u1s.resize(nscales_);
188         u2s.resize(nscales_);
189         u3s.resize(nscales_);
190 
191         I0.convertTo(I0s[0], CV_32F, I0.depth() == CV_8U ? 1.0 : 255.0, stream);
192         I1.convertTo(I1s[0], CV_32F, I1.depth() == CV_8U ? 1.0 : 255.0, stream);
193 
194         if (!useInitialFlow_)
195         {
196             flowx.create(I0.size(), CV_32FC1);
197             flowy.create(I0.size(), CV_32FC1);
198         }
199 
200         u1s[0] = flowx;
201         u2s[0] = flowy;
202         if (gamma_)
203         {
204             u3s[0].create(I0.size(), CV_32FC1);
205         }
206 
207         I1x_buf.create(I0.size(), CV_32FC1);
208         I1y_buf.create(I0.size(), CV_32FC1);
209 
210         I1w_buf.create(I0.size(), CV_32FC1);
211         I1wx_buf.create(I0.size(), CV_32FC1);
212         I1wy_buf.create(I0.size(), CV_32FC1);
213 
214         grad_buf.create(I0.size(), CV_32FC1);
215         rho_c_buf.create(I0.size(), CV_32FC1);
216 
217         p11_buf.create(I0.size(), CV_32FC1);
218         p12_buf.create(I0.size(), CV_32FC1);
219         p21_buf.create(I0.size(), CV_32FC1);
220         p22_buf.create(I0.size(), CV_32FC1);
221         if (gamma_)
222         {
223             p31_buf.create(I0.size(), CV_32FC1);
224             p32_buf.create(I0.size(), CV_32FC1);
225         }
226         diff_buf.create(I0.size(), CV_32FC1);
227 
228         // create the scales
229         for (int s = 1; s < nscales_; ++s)
230         {
231             cuda::resize(I0s[s-1], I0s[s], Size(), scaleStep_, scaleStep_, INTER_LINEAR, stream);
232             cuda::resize(I1s[s-1], I1s[s], Size(), scaleStep_, scaleStep_, INTER_LINEAR, stream);
233 
234             if (I0s[s].cols < 16 || I0s[s].rows < 16)
235             {
236                 nscales_ = s;
237                 break;
238             }
239 
240             if (useInitialFlow_)
241             {
242                 cuda::resize(u1s[s-1], u1s[s], Size(), scaleStep_, scaleStep_, INTER_LINEAR, stream);
243                 cuda::resize(u2s[s-1], u2s[s], Size(), scaleStep_, scaleStep_, INTER_LINEAR, stream);
244 
245                 cuda::multiply(u1s[s], Scalar::all(scaleStep_), u1s[s], 1, -1, stream);
246                 cuda::multiply(u2s[s], Scalar::all(scaleStep_), u2s[s], 1, -1, stream);
247             }
248             else
249             {
250                 u1s[s].create(I0s[s].size(), CV_32FC1);
251                 u2s[s].create(I0s[s].size(), CV_32FC1);
252             }
253             if (gamma_)
254             {
255                 u3s[s].create(I0s[s].size(), CV_32FC1);
256             }
257         }
258 
259         if (!useInitialFlow_)
260         {
261             u1s[nscales_-1].setTo(Scalar::all(0), stream);
262             u2s[nscales_-1].setTo(Scalar::all(0), stream);
263         }
264         if (gamma_)
265         {
266             u3s[nscales_ - 1].setTo(Scalar::all(0), stream);
267         }
268 
269         // pyramidal structure for computing the optical flow
270         for (int s = nscales_ - 1; s >= 0; --s)
271         {
272             // compute the optical flow at the current scale
273             procOneScale(I0s[s], I1s[s], u1s[s], u2s[s], u3s[s], stream);
274 
275             // if this was the last scale, finish now
276             if (s == 0)
277                 break;
278 
279             // otherwise, upsample the optical flow
280 
281             // zoom the optical flow for the next finer scale
282             cuda::resize(u1s[s], u1s[s - 1], I0s[s - 1].size(), 0, 0, INTER_LINEAR, stream);
283             cuda::resize(u2s[s], u2s[s - 1], I0s[s - 1].size(), 0, 0, INTER_LINEAR, stream);
284             if (gamma_)
285             {
286                 cuda::resize(u3s[s], u3s[s - 1], I0s[s - 1].size(), 0, 0, INTER_LINEAR, stream);
287             }
288 
289             // scale the optical flow with the appropriate zoom factor
290             cuda::multiply(u1s[s - 1], Scalar::all(1/scaleStep_), u1s[s - 1], 1, -1, stream);
291             cuda::multiply(u2s[s - 1], Scalar::all(1/scaleStep_), u2s[s - 1], 1, -1, stream);
292         }
293     }
294 
procOneScale(const GpuMat & I0,const GpuMat & I1,GpuMat & u1,GpuMat & u2,GpuMat & u3,Stream & _stream)295     void OpticalFlowDual_TVL1_Impl::procOneScale(const GpuMat& I0, const GpuMat& I1, GpuMat& u1, GpuMat& u2, GpuMat& u3, Stream& _stream)
296     {
297         using namespace tvl1flow;
298 
299         cudaStream_t stream = StreamAccessor::getStream(_stream);
300 
301         const double scaledEpsilon = epsilon_ * epsilon_ * I0.size().area();
302 
303         CV_DbgAssert( I1.size() == I0.size() );
304         CV_DbgAssert( I1.type() == I0.type() );
305         CV_DbgAssert( u1.size() == I0.size() );
306         CV_DbgAssert( u2.size() == u1.size() );
307 
308         GpuMat I1x = I1x_buf(Rect(0, 0, I0.cols, I0.rows));
309         GpuMat I1y = I1y_buf(Rect(0, 0, I0.cols, I0.rows));
310         centeredGradient(I1, I1x, I1y, stream);
311 
312         GpuMat I1w = I1w_buf(Rect(0, 0, I0.cols, I0.rows));
313         GpuMat I1wx = I1wx_buf(Rect(0, 0, I0.cols, I0.rows));
314         GpuMat I1wy = I1wy_buf(Rect(0, 0, I0.cols, I0.rows));
315 
316         GpuMat grad = grad_buf(Rect(0, 0, I0.cols, I0.rows));
317         GpuMat rho_c = rho_c_buf(Rect(0, 0, I0.cols, I0.rows));
318 
319         GpuMat p11 = p11_buf(Rect(0, 0, I0.cols, I0.rows));
320         GpuMat p12 = p12_buf(Rect(0, 0, I0.cols, I0.rows));
321         GpuMat p21 = p21_buf(Rect(0, 0, I0.cols, I0.rows));
322         GpuMat p22 = p22_buf(Rect(0, 0, I0.cols, I0.rows));
323         GpuMat p31, p32;
324         if (gamma_)
325         {
326             p31 = p31_buf(Rect(0, 0, I0.cols, I0.rows));
327             p32 = p32_buf(Rect(0, 0, I0.cols, I0.rows));
328         }
329         p11.setTo(Scalar::all(0), _stream);
330         p12.setTo(Scalar::all(0), _stream);
331         p21.setTo(Scalar::all(0), _stream);
332         p22.setTo(Scalar::all(0), _stream);
333         if (gamma_)
334         {
335             p31.setTo(Scalar::all(0), _stream);
336             p32.setTo(Scalar::all(0), _stream);
337         }
338 
339         GpuMat diff = diff_buf(Rect(0, 0, I0.cols, I0.rows));
340 
341         const float l_t = static_cast<float>(lambda_ * theta_);
342         const float taut = static_cast<float>(tau_ / theta_);
343 
344         for (int warpings = 0; warpings < warps_; ++warpings)
345         {
346             warpBackward(I0, I1, I1x, I1y, u1, u2, I1w, I1wx, I1wy, grad, rho_c, stream);
347 
348             double error = std::numeric_limits<double>::max();
349             double prevError = 0.0;
350             for (int n = 0; error > scaledEpsilon && n < iterations_; ++n)
351             {
352                 // some tweaks to make sum operation less frequently
353                 bool calcError = (epsilon_ > 0) && (n & 0x1) && (prevError < scaledEpsilon);
354                 estimateU(I1wx, I1wy, grad, rho_c, p11, p12, p21, p22, p31, p32, u1, u2, u3, diff, l_t, static_cast<float>(theta_), gamma_, calcError, stream);
355                 if (calcError)
356                 {
357                     _stream.waitForCompletion();
358                     error = cuda::sum(diff, norm_buf)[0];
359                     prevError = error;
360                 }
361                 else
362                 {
363                     error = std::numeric_limits<double>::max();
364                     prevError -= scaledEpsilon;
365                 }
366 
367                 estimateDualVariables(u1, u2, u3, p11, p12, p21, p22, p31, p32, taut, gamma_, stream);
368             }
369         }
370     }
371 }
372 
create(double tau,double lambda,double theta,int nscales,int warps,double epsilon,int iterations,double scaleStep,double gamma,bool useInitialFlow)373 Ptr<OpticalFlowDual_TVL1> cv::cuda::OpticalFlowDual_TVL1::create(
374             double tau, double lambda, double theta, int nscales, int warps,
375             double epsilon, int iterations, double scaleStep, double gamma, bool useInitialFlow)
376 {
377     return makePtr<OpticalFlowDual_TVL1_Impl>(tau, lambda, theta, nscales, warps,
378                                               epsilon, iterations, scaleStep, gamma, useInitialFlow);
379 }
380 
381 #endif // !defined HAVE_CUDA || defined(CUDA_DISABLER)
382