Commit 12ffd7f0 authored by jeffeDurand's avatar jeffeDurand Committed by Alexander Alekhin
Browse files

[moved from opencv] Merge pull request opencv/opencv#16090 from jeffeDurand:cuda_mog2_issue_5296 

* cuda_mog2_issue_5296
original commit: https://github.com/opencv/opencv/commit/5bf73457431b7d2cb87ac8c107865388dbf66642
parent f9bbe706

modules/cudabgsegm/src/cuda/mog2.cu

+280 −301
Original line number Diff line number Diff line
@@ -47,7 +47,13 @@ 
#include "opencv2/core/cuda/vec_math.hpp"

#include "opencv2/core/cuda/limits.hpp"



namespace cv { namespace cuda { namespace device

#include "mog2.hpp"



namespace cv

{

namespace cuda

{

namespace device

{

namespace mog2

{
@@ -104,43 +110,16 @@ namespace cv { namespace cuda { namespace device 
///////////////////////////////////////////////////////////////

// MOG2



        __constant__ int           c_nmixtures;

        __constant__ float         c_Tb;

        __constant__ float         c_TB;

        __constant__ float         c_Tg;

        __constant__ float         c_varInit;

        __constant__ float         c_varMin;

        __constant__ float         c_varMax;

        __constant__ float         c_tau;

        __constant__ unsigned char c_shadowVal;



        void loadConstants(int nmixtures, float Tb, float TB, float Tg, float varInit, float varMin, float varMax, float tau, unsigned char shadowVal)

        {

            varMin = ::fminf(varMin, varMax);

            varMax = ::fmaxf(varMin, varMax);



            cudaSafeCall( cudaMemcpyToSymbol(c_nmixtures, &nmixtures, sizeof(int)) );

            cudaSafeCall( cudaMemcpyToSymbol(c_Tb, &Tb, sizeof(float)) );

            cudaSafeCall( cudaMemcpyToSymbol(c_TB, &TB, sizeof(float)) );

            cudaSafeCall( cudaMemcpyToSymbol(c_Tg, &Tg, sizeof(float)) );

            cudaSafeCall( cudaMemcpyToSymbol(c_varInit, &varInit, sizeof(float)) );

            cudaSafeCall( cudaMemcpyToSymbol(c_varMin, &varMin, sizeof(float)) );

            cudaSafeCall( cudaMemcpyToSymbol(c_varMax, &varMax, sizeof(float)) );

            cudaSafeCall( cudaMemcpyToSymbol(c_tau, &tau, sizeof(float)) );

            cudaSafeCall( cudaMemcpyToSymbol(c_shadowVal, &shadowVal, sizeof(unsigned char)) );

        }



template <bool detectShadows, typename SrcT, typename WorkT>

__global__ void mog2(const PtrStepSz<SrcT> frame, PtrStepb fgmask, PtrStepb modesUsed,

                     PtrStepf gmm_weight, PtrStepf gmm_variance, PtrStep<WorkT> gmm_mean,

                             const float alphaT, const float alpha1, const float prune)

                     const float alphaT, const float alpha1, const float prune, const Constants *const constants)

{

    const int x = blockIdx.x * blockDim.x + threadIdx.x;

    const int y = blockIdx.y * blockDim.y + threadIdx.y;



            if (x >= frame.cols || y >= frame.rows)

                return;



    if (x < frame.cols && y < frame.rows)

    {

        WorkT pix = cvt(frame(y, x));



        //calculate distances to the modes (+ sort)
@@ -153,7 +132,7 @@ namespace cv { namespace cuda { namespace device 
        bool fitsPDF = false; //if it remains zero a new GMM mode will be added



        int nmodes = modesUsed(y, x);

            int nNewModes = nmodes; //current number of modes in GMM

        const int nNewModes = nmodes; //current number of modes in GMM



        float totalWeight = 0.0f;
@@ -168,20 +147,20 @@ namespace cv { namespace cuda { namespace device 
            if (!fitsPDF)

            {

                //check if it belongs to some of the remaining modes

                    float var = gmm_variance(mode * frame.rows + y, x);

                const float var = gmm_variance(mode * frame.rows + y, x);



                    WorkT mean = gmm_mean(mode * frame.rows + y, x);

                const WorkT mean = gmm_mean(mode * frame.rows + y, x);



                //calculate difference and distance

                    WorkT diff = mean - pix;

                    float dist2 = sqr(diff);

                const WorkT diff = mean - pix;

                const float dist2 = sqr(diff);



                //background? - Tb - usually larger than Tg

                    if (totalWeight < c_TB && dist2 < c_Tb * var)

                if (totalWeight < constants->TB_ && dist2 < constants->Tb_ * var)

                    background = true;



                //check fit

                    if (dist2 < c_Tg * var)

                if (dist2 < constants->Tg_ * var)

                {

                    //belongs to the mode

                    fitsPDF = true;
@@ -199,8 +178,8 @@ namespace cv { namespace cuda { namespace device 
                    float varnew = var + k * (dist2 - var);



                    //limit the variance

                        varnew = ::fmaxf(varnew, c_varMin);

                        varnew = ::fminf(varnew, c_varMax);

                    varnew = ::fmaxf(varnew, constants->varMin_);

                    varnew = ::fminf(varnew, constants->varMax_);



                    gmm_variance(mode * frame.rows + y, x) = varnew;
@@ -249,7 +228,7 @@ namespace cv { namespace cuda { namespace device 
        if (!fitsPDF)

        {

            // replace the weakest or add a new one

                int mode = nmodes == c_nmixtures ? c_nmixtures - 1 : nmodes++;

            const int mode = nmodes == constants->nmixtures_ ? constants->nmixtures_ - 1 : nmodes++;



            if (nmodes == 1)

                gmm_weight(mode * frame.rows + y, x) = 1.f;
@@ -266,7 +245,7 @@ namespace cv { namespace cuda { namespace device 
            // init



            gmm_mean(mode * frame.rows + y, x) = pix;

                gmm_variance(mode * frame.rows + y, x) = c_varInit;

            gmm_variance(mode * frame.rows + y, x) = constants->varInit_;



            //sort

            //find the new place for it
@@ -295,25 +274,25 @@ namespace cv { namespace cuda { namespace device 
            // check all the components  marked as background:

            for (int mode = 0; mode < nmodes; ++mode)

            {

                    WorkT mean = gmm_mean(mode * frame.rows + y, x);

                const WorkT mean = gmm_mean(mode * frame.rows + y, x);



                    WorkT pix_mean = pix * mean;

                const WorkT pix_mean = pix * mean;



                    float numerator = sum(pix_mean);

                    float denominator = sqr(mean);

                const float numerator = sum(pix_mean);

                const float denominator = sqr(mean);



                // no division by zero allowed

                if (denominator == 0)

                    break;



                // if tau < a < 1 then also check the color distortion

                    if (numerator <= denominator && numerator >= c_tau * denominator)

                else if (numerator <= denominator && numerator >= constants->tau_ * denominator)

                {

                        float a = numerator / denominator;

                    const float a = numerator / denominator;



                    WorkT dD = a * mean - pix;



                        if (sqr(dD) < c_Tb * gmm_variance(mode * frame.rows + y, x) * a * a)

                    if (sqr(dD) < constants->Tb_ * gmm_variance(mode * frame.rows + y, x) * a * a)

                    {

                        isShadow = true;

                        break;
@@ -321,17 +300,18 @@ namespace cv { namespace cuda { namespace device 
                };



                tWeight += gmm_weight(mode * frame.rows + y, x);

                    if (tWeight > c_TB)

                if (tWeight > constants->TB_)

                    break;

            }

        }



            fgmask(y, x) = background ? 0 : isShadow ? c_shadowVal : 255;

        fgmask(y, x) = background ? 0 : isShadow ? constants->shadowVal_ : 255;

    }

}



template <typename SrcT, typename WorkT>

void mog2_caller(PtrStepSzb frame, PtrStepSzb fgmask, PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzf variance, PtrStepSzb mean,

                         float alphaT, float prune, bool detectShadows, cudaStream_t stream)

                 float alphaT, float prune, bool detectShadows, const Constants *const constants, cudaStream_t stream)

{

    dim3 block(32, 8);

    dim3 grid(divUp(frame.cols, block.x), divUp(frame.rows, block.y));
@@ -344,7 +324,7 @@ namespace cv { namespace cuda { namespace device 


        mog2<true, SrcT, WorkT><<<grid, block, 0, stream>>>((PtrStepSz<SrcT>)frame, fgmask, modesUsed,

                                                            weight, variance, (PtrStepSz<WorkT>)mean,

                                                                    alphaT, alpha1, prune);

                                                            alphaT, alpha1, prune, constants);

    }

    else

    {
@@ -352,7 +332,7 @@ namespace cv { namespace cuda { namespace device 


        mog2<false, SrcT, WorkT><<<grid, block, 0, stream>>>((PtrStepSz<SrcT>)frame, fgmask, modesUsed,

                                                             weight, variance, (PtrStepSz<WorkT>)mean,

                                                                    alphaT, alpha1, prune);

                                                             alphaT, alpha1, prune, constants);

    }



    cudaSafeCall(cudaGetLastError());
@@ -362,20 +342,19 @@ namespace cv { namespace cuda { namespace device 
}



void mog2_gpu(PtrStepSzb frame, int cn, PtrStepSzb fgmask, PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzf variance, PtrStepSzb mean,

                      float alphaT, float prune, bool detectShadows, cudaStream_t stream)

              float alphaT, float prune, bool detectShadows, const Constants *const constants, cudaStream_t stream)

{

            typedef void (*func_t)(PtrStepSzb frame, PtrStepSzb fgmask, PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzf variance, PtrStepSzb mean, float alphaT, float prune, bool detectShadows, cudaStream_t stream);

    typedef void (*func_t)(PtrStepSzb frame, PtrStepSzb fgmask, PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzf variance, PtrStepSzb mean, float alphaT, float prune, bool detectShadows, const Constants *const constants, cudaStream_t stream);



    static const func_t funcs[] =

        {

                0, mog2_caller<uchar, float>, 0, mog2_caller<uchar3, float3>, mog2_caller<uchar4, float4>

            };

            0, mog2_caller<uchar, float>, 0, mog2_caller<uchar3, float3>, mog2_caller<uchar4, float4>};



            funcs[cn](frame, fgmask, modesUsed, weight, variance, mean, alphaT, prune, detectShadows, stream);

    funcs[cn](frame, fgmask, modesUsed, weight, variance, mean, alphaT, prune, detectShadows, constants, stream);

}



template <typename WorkT, typename OutT>

        __global__ void getBackgroundImage2(const PtrStepSzb modesUsed, const PtrStepf gmm_weight, const PtrStep<WorkT> gmm_mean, PtrStep<OutT> dst)

__global__ void getBackgroundImage2(const PtrStepSzb modesUsed, const PtrStepf gmm_weight, const PtrStep<WorkT> gmm_mean, PtrStep<OutT> dst, const Constants *const constants)

{

    const int x = blockIdx.x * blockDim.x + threadIdx.x;

    const int y = blockIdx.y * blockDim.y + threadIdx.y;
@@ -397,7 +376,7 @@ namespace cv { namespace cuda { namespace device 


        totalWeight += weight;



                if(totalWeight > c_TB)

        if (totalWeight > constants->TB_)

            break;

    }
@@ -407,33 +386,33 @@ namespace cv { namespace cuda { namespace device 
}



template <typename WorkT, typename OutT>

        void getBackgroundImage2_caller(PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzb mean, PtrStepSzb dst, cudaStream_t stream)

void getBackgroundImage2_caller(PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzb mean, PtrStepSzb dst, const Constants *const constants, cudaStream_t stream)

{

    dim3 block(32, 8);

    dim3 grid(divUp(modesUsed.cols, block.x), divUp(modesUsed.rows, block.y));



    cudaSafeCall(cudaFuncSetCacheConfig(getBackgroundImage2<WorkT, OutT>, cudaFuncCachePreferL1));



            getBackgroundImage2<WorkT, OutT><<<grid, block, 0, stream>>>(modesUsed, weight, (PtrStepSz<WorkT>) mean, (PtrStepSz<OutT>) dst);

    getBackgroundImage2<WorkT, OutT><<<grid, block, 0, stream>>>(modesUsed, weight, (PtrStepSz<WorkT>)mean, (PtrStepSz<OutT>)dst, constants);

    cudaSafeCall(cudaGetLastError());



    if (stream == 0)

        cudaSafeCall(cudaDeviceSynchronize());

}



        void getBackgroundImage2_gpu(int cn, PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzb mean, PtrStepSzb dst, cudaStream_t stream)

void getBackgroundImage2_gpu(int cn, PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzb mean, PtrStepSzb dst, const Constants *const constants, cudaStream_t stream)

{

            typedef void (*func_t)(PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzb mean, PtrStepSzb dst, cudaStream_t stream);

    typedef void (*func_t)(PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzb mean, PtrStepSzb dst, const Constants *const constants, cudaStream_t stream);



    static const func_t funcs[] =

        {

                0, getBackgroundImage2_caller<float, uchar>, 0, getBackgroundImage2_caller<float3, uchar3>, getBackgroundImage2_caller<float4, uchar4>

            };

            0, getBackgroundImage2_caller<float, uchar>, 0, getBackgroundImage2_caller<float3, uchar3>, getBackgroundImage2_caller<float4, uchar4>};



            funcs[cn](modesUsed, weight, mean, dst, stream);

        }

    funcs[cn](modesUsed, weight, mean, dst, constants, stream);

}

}}}



} // namespace mog2

} // namespace device

} // namespace cuda

} // namespace cv



#endif /* CUDA_DISABLER */

modules/cudabgsegm/src/cuda/mog2.hpp

0 → 100644
+37 −0
Original line number Diff line number Diff line 
// This file is part of OpenCV project.

// It is subject to the license terms in the LICENSE file found in the top-level directory

// of this distribution and at http://opencv.org/license.html.



#ifndef OPENCV_CUDA_MOG2_H

#define OPENCV_CUDA_MOG2_H



#include "opencv2/core/cuda.hpp"



struct CUstream_st;

typedef struct CUstream_st *cudaStream_t;



namespace cv { namespace cuda {



class Stream;



namespace device { namespace mog2 {



typedef struct

{

    float Tb_;

    float TB_;

    float Tg_;

    float varInit_;

    float varMin_;

    float varMax_;

    float tau_;

    int nmixtures_;

    unsigned char shadowVal_;

} Constants;



void mog2_gpu(PtrStepSzb frame, int cn, PtrStepSzb fgmask, PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzf variance, PtrStepSzb mean, float alphaT, float prune, bool detectShadows, const Constants *const constants, cudaStream_t stream);

void getBackgroundImage2_gpu(int cn, PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzb mean, PtrStepSzb dst, const Constants *const constants, cudaStream_t stream);



} } } }



#endif /* OPENCV_CUDA_MOG2_H */

modules/cudabgsegm/src/mog2.cpp

+176 −178
Original line number Diff line number Diff line
@@ -41,25 +41,21 @@ 
//M*/



#include "precomp.hpp"

#include "cuda/mog2.hpp"



using namespace cv;

using namespace cv::cuda;

using namespace cv::cuda::device::mog2;



#if !defined HAVE_CUDA || defined(CUDA_DISABLER)



Ptr<cuda::BackgroundSubtractorMOG2> cv::cuda::createBackgroundSubtractorMOG2(int, double, bool) { throw_no_cuda(); return Ptr<cuda::BackgroundSubtractorMOG2>(); }



#else



namespace cv { namespace cuda { namespace device

{

    namespace mog2

Ptr<cuda::BackgroundSubtractorMOG2> cv::cuda::createBackgroundSubtractorMOG2(int, double, bool)

{

        void loadConstants(int nmixtures, float Tb, float TB, float Tg, float varInit, float varMin, float varMax, float tau, unsigned char shadowVal);

        void mog2_gpu(PtrStepSzb frame, int cn, PtrStepSzb fgmask, PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzf variance, PtrStepSzb mean, float alphaT, float prune, bool detectShadows, cudaStream_t stream);

        void getBackgroundImage2_gpu(int cn, PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzb mean, PtrStepSzb dst, cudaStream_t stream);

    throw_no_cuda();

    return Ptr<cuda::BackgroundSubtractorMOG2>();

}

}}}



#else



namespace

{
@@ -82,6 +78,7 @@ namespace 
{

public:

    MOG2Impl(int history, double varThreshold, bool detectShadows);

    ~MOG2Impl();



    void apply(InputArray image, OutputArray fgmask, double learningRate = -1) CV_OVERRIDE;

    void apply(InputArray image, OutputArray fgmask, double learningRate, Stream &stream) CV_OVERRIDE;
@@ -92,26 +89,26 @@ namespace 
    int getHistory() const CV_OVERRIDE { return history_; }

    void setHistory(int history) CV_OVERRIDE { history_ = history; }



        int getNMixtures() const CV_OVERRIDE { return nmixtures_; }

        void setNMixtures(int nmixtures) CV_OVERRIDE { nmixtures_ = nmixtures; }

    int getNMixtures() const CV_OVERRIDE { return constantsHost_.nmixtures_; }

    void setNMixtures(int nmixtures) CV_OVERRIDE { constantsHost_.nmixtures_ = nmixtures; }



        double getBackgroundRatio() const CV_OVERRIDE { return backgroundRatio_; }

        void setBackgroundRatio(double ratio) CV_OVERRIDE { backgroundRatio_ = (float) ratio; }

    double getBackgroundRatio() const CV_OVERRIDE { return constantsHost_.TB_; }

    void setBackgroundRatio(double ratio) CV_OVERRIDE { constantsHost_.TB_ = (float)ratio; }



        double getVarThreshold() const CV_OVERRIDE { return varThreshold_; }

        void setVarThreshold(double varThreshold) CV_OVERRIDE { varThreshold_ = (float) varThreshold; }

    double getVarThreshold() const CV_OVERRIDE { return constantsHost_.Tb_; }

    void setVarThreshold(double varThreshold) CV_OVERRIDE { constantsHost_.Tb_ = (float)varThreshold; }



        double getVarThresholdGen() const CV_OVERRIDE { return varThresholdGen_; }

        void setVarThresholdGen(double varThresholdGen) CV_OVERRIDE { varThresholdGen_ = (float) varThresholdGen; }

    double getVarThresholdGen() const CV_OVERRIDE { return constantsHost_.Tg_; }

    void setVarThresholdGen(double varThresholdGen) CV_OVERRIDE { constantsHost_.Tg_ = (float)varThresholdGen; }



        double getVarInit() const CV_OVERRIDE { return varInit_; }

        void setVarInit(double varInit) CV_OVERRIDE { varInit_ = (float) varInit; }

    double getVarInit() const CV_OVERRIDE { return constantsHost_.varInit_; }

    void setVarInit(double varInit) CV_OVERRIDE { constantsHost_.varInit_ = (float)varInit; }



        double getVarMin() const CV_OVERRIDE { return varMin_; }

        void setVarMin(double varMin) CV_OVERRIDE { varMin_ = (float) varMin; }

    double getVarMin() const CV_OVERRIDE { return constantsHost_.varMin_; }

    void setVarMin(double varMin) CV_OVERRIDE { constantsHost_.varMin_ = ::fminf((float)varMin, constantsHost_.varMax_); }



        double getVarMax() const CV_OVERRIDE { return varMax_; }

        void setVarMax(double varMax) CV_OVERRIDE { varMax_ = (float) varMax; }

    double getVarMax() const CV_OVERRIDE { return constantsHost_.varMax_; }

    void setVarMax(double varMax) CV_OVERRIDE { constantsHost_.varMax_ = ::fmaxf(constantsHost_.varMin_, (float)varMax); }



    double getComplexityReductionThreshold() const CV_OVERRIDE { return ct_; }

    void setComplexityReductionThreshold(double ct) CV_OVERRIDE { ct_ = (float)ct; }
@@ -119,27 +116,21 @@ namespace 
    bool getDetectShadows() const CV_OVERRIDE { return detectShadows_; }

    void setDetectShadows(bool detectShadows) CV_OVERRIDE { detectShadows_ = detectShadows; }



        int getShadowValue() const CV_OVERRIDE { return shadowValue_; }

        void setShadowValue(int value) CV_OVERRIDE { shadowValue_ = (uchar) value; }

    int getShadowValue() const CV_OVERRIDE { return constantsHost_.shadowVal_; }

    void setShadowValue(int value) CV_OVERRIDE { constantsHost_.shadowVal_ = (uchar)value; }



        double getShadowThreshold() const CV_OVERRIDE { return shadowThreshold_; }

        void setShadowThreshold(double threshold) CV_OVERRIDE { shadowThreshold_ = (float) threshold; }

    double getShadowThreshold() const CV_OVERRIDE { return constantsHost_.tau_; }

    void setShadowThreshold(double threshold) CV_OVERRIDE { constantsHost_.tau_ = (float)threshold; }



private:

        void initialize(Size frameSize, int frameType);

    void initialize(Size frameSize, int frameType, Stream &stream);



    Constants constantsHost_;

    Constants *constantsDevice_;



    int history_;

        int nmixtures_;

        float backgroundRatio_;

        float varThreshold_;

        float varThresholdGen_;

        float varInit_;

        float varMin_;

        float varMax_;

    float ct_;

    bool detectShadows_;

        uchar shadowValue_;

        float shadowThreshold_;



    Size frameSize_;

    int frameType_;
@@ -153,22 +144,29 @@ namespace 
    GpuMat bgmodelUsedModes_;

};



    MOG2Impl::MOG2Impl(int history, double varThreshold, bool detectShadows) :

        frameSize_(0, 0), frameType_(0), nframes_(0)

MOG2Impl::MOG2Impl(int history, double varThreshold, bool detectShadows) : frameSize_(0, 0), frameType_(0), nframes_(0)

{

    history_ = history > 0 ? history : defaultHistory;

        varThreshold_ = varThreshold > 0 ? (float) varThreshold : defaultVarThreshold;

    detectShadows_ = detectShadows;



        nmixtures_ = defaultNMixtures;

        backgroundRatio_ = defaultBackgroundRatio;

        varInit_ = defaultVarInit;

        varMax_ = defaultVarMax;

        varMin_ = defaultVarMin;

        varThresholdGen_ = defaultVarThresholdGen;

    ct_ = defaultCT;

        shadowValue_ =  defaultShadowValue;

        shadowThreshold_ = defaultShadowThreshold;



    setNMixtures(defaultNMixtures);

    setBackgroundRatio(defaultBackgroundRatio);

    setVarInit(defaultVarInit);

    setVarMin(defaultVarMin);

    setVarMax(defaultVarMax);

    setVarThreshold(varThreshold > 0 ? (float)varThreshold : defaultVarThreshold);

    setVarThresholdGen(defaultVarThresholdGen);



    setShadowValue(defaultShadowValue);

    setShadowThreshold(defaultShadowThreshold);



    cudaSafeCall(cudaMalloc((void **)&constantsDevice_, sizeof(Constants)));

}



MOG2Impl::~MOG2Impl()

{

    cudaFree(constantsDevice_);

}



void MOG2Impl::apply(InputArray image, OutputArray fgmask, double learningRate)
@@ -186,7 +184,7 @@ namespace 
    int work_ch = ch;



    if (nframes_ == 0 || learningRate >= 1.0 || frame.size() != frameSize_ || work_ch != mean_.channels())

            initialize(frame.size(), frame.type());

        initialize(frame.size(), frame.type(), stream);



    _fgmask.create(frameSize_, CV_8UC1);

    GpuMat fgmask = _fgmask.getGpuMat();
@@ -198,7 +196,7 @@ namespace 
    CV_Assert(learningRate >= 0);



    mog2_gpu(frame, frame.channels(), fgmask, bgmodelUsedModes_, weight_, variance_, mean_,

                 (float) learningRate, static_cast<float>(-learningRate * ct_), detectShadows_, StreamAccessor::getStream(stream));

             (float)learningRate, static_cast<float>(-learningRate * ct_), detectShadows_, constantsDevice_, StreamAccessor::getStream(stream));

}



void MOG2Impl::getBackgroundImage(OutputArray backgroundImage) const
@@ -213,10 +211,10 @@ namespace 
    _backgroundImage.create(frameSize_, frameType_);

    GpuMat backgroundImage = _backgroundImage.getGpuMat();



        getBackgroundImage2_gpu(backgroundImage.channels(), bgmodelUsedModes_, weight_, mean_, backgroundImage, StreamAccessor::getStream(stream));

    getBackgroundImage2_gpu(backgroundImage.channels(), bgmodelUsedModes_, weight_, mean_, backgroundImage, constantsDevice_, StreamAccessor::getStream(stream));

}



    void MOG2Impl::initialize(cv::Size frameSize, int frameType)

void MOG2Impl::initialize(cv::Size frameSize, int frameType, Stream &stream)

{

    using namespace cv::cuda::device::mog2;
@@ -226,24 +224,24 @@ namespace 
    frameType_ = frameType;

    nframes_ = 0;



        int ch = CV_MAT_CN(frameType);

        int work_ch = ch;

    const int ch = CV_MAT_CN(frameType);

    const int work_ch = ch;



    // for each gaussian mixture of each pixel bg model we store ...

    // the mixture weight (w),

    // the mean (nchannels values) and

    // the covariance

        weight_.create(frameSize.height * nmixtures_, frameSize_.width, CV_32FC1);

        variance_.create(frameSize.height * nmixtures_, frameSize_.width, CV_32FC1);

        mean_.create(frameSize.height * nmixtures_, frameSize_.width, CV_32FC(work_ch));

    weight_.create(frameSize.height * getNMixtures(), frameSize_.width, CV_32FC1);

    variance_.create(frameSize.height * getNMixtures(), frameSize_.width, CV_32FC1);

    mean_.create(frameSize.height * getNMixtures(), frameSize_.width, CV_32FC(work_ch));



    //make the array for keeping track of the used modes per pixel - all zeros at start

    bgmodelUsedModes_.create(frameSize_, CV_8UC1);

    bgmodelUsedModes_.setTo(Scalar::all(0));



        loadConstants(nmixtures_, varThreshold_, backgroundRatio_, varThresholdGen_, varInit_, varMin_, varMax_, shadowThreshold_, shadowValue_);

    }

    cudaSafeCall(cudaMemcpyAsync(constantsDevice_, &constantsHost_, sizeof(Constants), cudaMemcpyHostToDevice, StreamAccessor::getStream(stream)));

}

} // namespace



Ptr<cuda::BackgroundSubtractorMOG2> cv::cuda::createBackgroundSubtractorMOG2(int history, double varThreshold, bool detectShadows)

{