Merge pull request #16090 from jeffeDurand:cuda_mog2_issue_5296

* cuda_mog2_issue_5296
pull/16200/head^2
jeffeDurand 5 years ago committed by Alexander Alekhin
parent 4342657762
commit 5bf7345743
  1. 581
      modules/cudabgsegm/src/cuda/mog2.cu
  2. 37
      modules/cudabgsegm/src/cuda/mog2.hpp
  3. 354
      modules/cudabgsegm/src/mog2.cpp

@ -47,393 +47,372 @@
#include "opencv2/core/cuda/vec_math.hpp" #include "opencv2/core/cuda/vec_math.hpp"
#include "opencv2/core/cuda/limits.hpp" #include "opencv2/core/cuda/limits.hpp"
namespace cv { namespace cuda { namespace device #include "mog2.hpp"
{
namespace mog2
{
///////////////////////////////////////////////////////////////
// Utility
__device__ __forceinline__ float cvt(uchar val)
{
return val;
}
__device__ __forceinline__ float3 cvt(const uchar3& val)
{
return make_float3(val.x, val.y, val.z);
}
__device__ __forceinline__ float4 cvt(const uchar4& val)
{
return make_float4(val.x, val.y, val.z, val.w);
}
__device__ __forceinline__ float sqr(float val)
{
return val * val;
}
__device__ __forceinline__ float sqr(const float3& val)
{
return val.x * val.x + val.y * val.y + val.z * val.z;
}
__device__ __forceinline__ float sqr(const float4& val)
{
return val.x * val.x + val.y * val.y + val.z * val.z;
}
__device__ __forceinline__ float sum(float val) namespace cv
{ {
return val; namespace cuda
} {
__device__ __forceinline__ float sum(const float3& val) namespace device
{ {
return val.x + val.y + val.z; namespace mog2
} {
__device__ __forceinline__ float sum(const float4& val) ///////////////////////////////////////////////////////////////
{ // Utility
return val.x + val.y + val.z;
}
template <class Ptr2D>
__device__ __forceinline__ void swap(Ptr2D& ptr, int x, int y, int k, int rows)
{
typename Ptr2D::elem_type val = ptr(k * rows + y, x);
ptr(k * rows + y, x) = ptr((k + 1) * rows + y, x);
ptr((k + 1) * rows + y, x) = val;
}
///////////////////////////////////////////////////////////////
// MOG2
__constant__ int c_nmixtures; __device__ __forceinline__ float cvt(uchar val)
__constant__ float c_Tb; {
__constant__ float c_TB; return val;
__constant__ float c_Tg; }
__constant__ float c_varInit; __device__ __forceinline__ float3 cvt(const uchar3 &val)
__constant__ float c_varMin; {
__constant__ float c_varMax; return make_float3(val.x, val.y, val.z);
__constant__ float c_tau; }
__constant__ unsigned char c_shadowVal; __device__ __forceinline__ float4 cvt(const uchar4 &val)
{
return make_float4(val.x, val.y, val.z, val.w);
}
void loadConstants(int nmixtures, float Tb, float TB, float Tg, float varInit, float varMin, float varMax, float tau, unsigned char shadowVal) __device__ __forceinline__ float sqr(float val)
{ {
varMin = ::fminf(varMin, varMax); return val * val;
varMax = ::fmaxf(varMin, varMax); }
__device__ __forceinline__ float sqr(const float3 &val)
cudaSafeCall( cudaMemcpyToSymbol(c_nmixtures, &nmixtures, sizeof(int)) ); {
cudaSafeCall( cudaMemcpyToSymbol(c_Tb, &Tb, sizeof(float)) ); return val.x * val.x + val.y * val.y + val.z * val.z;
cudaSafeCall( cudaMemcpyToSymbol(c_TB, &TB, sizeof(float)) ); }
cudaSafeCall( cudaMemcpyToSymbol(c_Tg, &Tg, sizeof(float)) ); __device__ __forceinline__ float sqr(const float4 &val)
cudaSafeCall( cudaMemcpyToSymbol(c_varInit, &varInit, sizeof(float)) ); {
cudaSafeCall( cudaMemcpyToSymbol(c_varMin, &varMin, sizeof(float)) ); return val.x * val.x + val.y * val.y + val.z * val.z;
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> __device__ __forceinline__ float sum(float val)
__global__ void mog2(const PtrStepSz<SrcT> frame, PtrStepb fgmask, PtrStepb modesUsed, {
PtrStepf gmm_weight, PtrStepf gmm_variance, PtrStep<WorkT> gmm_mean, return val;
const float alphaT, const float alpha1, const float prune) }
{ __device__ __forceinline__ float sum(const float3 &val)
const int x = blockIdx.x * blockDim.x + threadIdx.x; {
const int y = blockIdx.y * blockDim.y + threadIdx.y; return val.x + val.y + val.z;
}
__device__ __forceinline__ float sum(const float4 &val)
{
return val.x + val.y + val.z;
}
if (x >= frame.cols || y >= frame.rows) template <class Ptr2D>
return; __device__ __forceinline__ void swap(Ptr2D &ptr, int x, int y, int k, int rows)
{
typename Ptr2D::elem_type val = ptr(k * rows + y, x);
ptr(k * rows + y, x) = ptr((k + 1) * rows + y, x);
ptr((k + 1) * rows + y, x) = val;
}
///////////////////////////////////////////////////////////////
// MOG2
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 Constants *const constants)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
WorkT pix = cvt(frame(y, x)); if (x < frame.cols && y < frame.rows)
{
WorkT pix = cvt(frame(y, x));
//calculate distances to the modes (+ sort) //calculate distances to the modes (+ sort)
//here we need to go in descending order!!! //here we need to go in descending order!!!
bool background = false; // true - the pixel classified as background bool background = false; // true - the pixel classified as background
//internal: //internal:
bool fitsPDF = false; //if it remains zero a new GMM mode will be added bool fitsPDF = false; //if it remains zero a new GMM mode will be added
int nmodes = modesUsed(y, x); 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; float totalWeight = 0.0f;
//go through all modes //go through all modes
for (int mode = 0; mode < nmodes; ++mode) for (int mode = 0; mode < nmodes; ++mode)
{
//need only weight if fit is found
float weight = alpha1 * gmm_weight(mode * frame.rows + y, x) + prune;
int swap_count = 0;
//fit not found yet
if (!fitsPDF)
{ {
//need only weight if fit is found //check if it belongs to some of the remaining modes
float weight = alpha1 * gmm_weight(mode * frame.rows + y, x) + prune; const float var = gmm_variance(mode * frame.rows + y, x);
int swap_count = 0;
//fit not found yet
if (!fitsPDF)
{
//check if it belongs to some of the remaining modes
float var = gmm_variance(mode * frame.rows + y, x);
WorkT mean = gmm_mean(mode * frame.rows + y, x);
//calculate difference and distance const WorkT mean = gmm_mean(mode * frame.rows + y, x);
WorkT diff = mean - pix;
float dist2 = sqr(diff);
//background? - Tb - usually larger than Tg //calculate difference and distance
if (totalWeight < c_TB && dist2 < c_Tb * var) const WorkT diff = mean - pix;
background = true; const float dist2 = sqr(diff);
//check fit //background? - Tb - usually larger than Tg
if (dist2 < c_Tg * var) if (totalWeight < constants->TB_ && dist2 < constants->Tb_ * var)
{ background = true;
//belongs to the mode
fitsPDF = true;
//update distribution //check fit
if (dist2 < constants->Tg_ * var)
{
//belongs to the mode
fitsPDF = true;
//update weight //update distribution
weight += alphaT;
float k = alphaT / weight;
//update mean //update weight
gmm_mean(mode * frame.rows + y, x) = mean - k * diff; weight += alphaT;
float k = alphaT / weight;
//update variance //update mean
float varnew = var + k * (dist2 - var); gmm_mean(mode * frame.rows + y, x) = mean - k * diff;
//limit the variance //update variance
varnew = ::fmaxf(varnew, c_varMin); float varnew = var + k * (dist2 - var);
varnew = ::fminf(varnew, c_varMax);
gmm_variance(mode * frame.rows + y, x) = varnew; //limit the variance
varnew = ::fmaxf(varnew, constants->varMin_);
varnew = ::fminf(varnew, constants->varMax_);
//sort gmm_variance(mode * frame.rows + y, x) = varnew;
//all other weights are at the same place and
//only the matched (iModes) is higher -> just find the new place for it
for (int i = mode; i > 0; --i) //sort
{ //all other weights are at the same place and
//check one up //only the matched (iModes) is higher -> just find the new place for it
if (weight < gmm_weight((i - 1) * frame.rows + y, x))
break;
swap_count++; for (int i = mode; i > 0; --i)
//swap one up {
swap(gmm_weight, x, y, i - 1, frame.rows); //check one up
swap(gmm_variance, x, y, i - 1, frame.rows); if (weight < gmm_weight((i - 1) * frame.rows + y, x))
swap(gmm_mean, x, y, i - 1, frame.rows); break;
}
//belongs to the mode - bFitsPDF becomes 1 swap_count++;
//swap one up
swap(gmm_weight, x, y, i - 1, frame.rows);
swap(gmm_variance, x, y, i - 1, frame.rows);
swap(gmm_mean, x, y, i - 1, frame.rows);
} }
} // !fitsPDF
//check prune //belongs to the mode - bFitsPDF becomes 1
if (weight < -prune)
{
weight = 0.0f;
nmodes--;
} }
} // !fitsPDF
gmm_weight((mode - swap_count) * frame.rows + y, x) = weight; //update weight by the calculated value //check prune
totalWeight += weight; if (weight < -prune)
{
weight = 0.0f;
nmodes--;
} }
//renormalize weights gmm_weight((mode - swap_count) * frame.rows + y, x) = weight; //update weight by the calculated value
totalWeight += weight;
}
totalWeight = 1.f / totalWeight; //renormalize weights
for (int mode = 0; mode < nmodes; ++mode)
gmm_weight(mode * frame.rows + y, x) *= totalWeight;
nmodes = nNewModes; totalWeight = 1.f / totalWeight;
for (int mode = 0; mode < nmodes; ++mode)
gmm_weight(mode * frame.rows + y, x) *= totalWeight;
//make new mode if needed and exit nmodes = nNewModes;
if (!fitsPDF) //make new mode if needed and exit
{
// replace the weakest or add a new one
int mode = nmodes == c_nmixtures ? c_nmixtures - 1 : nmodes++;
if (nmodes == 1) if (!fitsPDF)
gmm_weight(mode * frame.rows + y, x) = 1.f; {
else // replace the weakest or add a new one
{ const int mode = nmodes == constants->nmixtures_ ? constants->nmixtures_ - 1 : nmodes++;
gmm_weight(mode * frame.rows + y, x) = alphaT;
// renormalize all other weights if (nmodes == 1)
gmm_weight(mode * frame.rows + y, x) = 1.f;
else
{
gmm_weight(mode * frame.rows + y, x) = alphaT;
for (int i = 0; i < nmodes - 1; ++i) // renormalize all other weights
gmm_weight(i * frame.rows + y, x) *= alpha1;
}
// init for (int i = 0; i < nmodes - 1; ++i)
gmm_weight(i * frame.rows + y, x) *= alpha1;
}
gmm_mean(mode * frame.rows + y, x) = pix; // init
gmm_variance(mode * frame.rows + y, x) = c_varInit;
//sort gmm_mean(mode * frame.rows + y, x) = pix;
//find the new place for it gmm_variance(mode * frame.rows + y, x) = constants->varInit_;
for (int i = nmodes - 1; i > 0; --i) //sort
{ //find the new place for it
// check one up
if (alphaT < gmm_weight((i - 1) * frame.rows + y, x))
break;
//swap one up for (int i = nmodes - 1; i > 0; --i)
swap(gmm_weight, x, y, i - 1, frame.rows); {
swap(gmm_variance, x, y, i - 1, frame.rows); // check one up
swap(gmm_mean, x, y, i - 1, frame.rows); if (alphaT < gmm_weight((i - 1) * frame.rows + y, x))
} break;
//swap one up
swap(gmm_weight, x, y, i - 1, frame.rows);
swap(gmm_variance, x, y, i - 1, frame.rows);
swap(gmm_mean, x, y, i - 1, frame.rows);
} }
}
//set the number of modes //set the number of modes
modesUsed(y, x) = nmodes; modesUsed(y, x) = nmodes;
bool isShadow = false; bool isShadow = false;
if (detectShadows && !background) if (detectShadows && !background)
{ {
float tWeight = 0.0f; float tWeight = 0.0f;
// check all the components marked as background: // check all the components marked as background:
for (int mode = 0; mode < nmodes; ++mode) 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); const float numerator = sum(pix_mean);
float denominator = sqr(mean); const float denominator = sqr(mean);
// no division by zero allowed // no division by zero allowed
if (denominator == 0) if (denominator == 0)
break; break;
// if tau < a < 1 then also check the color distortion
if (numerator <= denominator && numerator >= c_tau * denominator)
{
float a = numerator / denominator;
WorkT dD = a * mean - pix; // if tau < a < 1 then also check the color distortion
else if (numerator <= denominator && numerator >= constants->tau_ * denominator)
{
const float a = numerator / denominator;
if (sqr(dD) < c_Tb * gmm_variance(mode * frame.rows + y, x) * a * a) WorkT dD = a * mean - pix;
{
isShadow = true;
break;
}
};
tWeight += gmm_weight(mode * frame.rows + y, x); if (sqr(dD) < constants->Tb_ * gmm_variance(mode * frame.rows + y, x) * a * a)
if (tWeight > c_TB) {
isShadow = true;
break; break;
} }
} };
fgmask(y, x) = background ? 0 : isShadow ? c_shadowVal : 255; tWeight += gmm_weight(mode * frame.rows + y, x);
if (tWeight > constants->TB_)
break;
}
} }
template <typename SrcT, typename WorkT> fgmask(y, x) = background ? 0 : isShadow ? constants->shadowVal_ : 255;
void mog2_caller(PtrStepSzb frame, PtrStepSzb fgmask, PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzf variance, PtrStepSzb mean, }
float alphaT, float prune, bool detectShadows, cudaStream_t stream) }
{
dim3 block(32, 8);
dim3 grid(divUp(frame.cols, block.x), divUp(frame.rows, block.y));
const float alpha1 = 1.0f - alphaT;
if (detectShadows) template <typename SrcT, typename WorkT>
{ void mog2_caller(PtrStepSzb frame, PtrStepSzb fgmask, PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzf variance, PtrStepSzb mean,
cudaSafeCall( cudaFuncSetCacheConfig(mog2<true, SrcT, WorkT>, cudaFuncCachePreferL1) ); 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));
mog2<true, SrcT, WorkT><<<grid, block, 0, stream>>>((PtrStepSz<SrcT>) frame, fgmask, modesUsed, const float alpha1 = 1.0f - alphaT;
weight, variance, (PtrStepSz<WorkT>) mean,
alphaT, alpha1, prune);
}
else
{
cudaSafeCall( cudaFuncSetCacheConfig(mog2<false, SrcT, WorkT>, cudaFuncCachePreferL1) );
mog2<false, SrcT, WorkT><<<grid, block, 0, stream>>>((PtrStepSz<SrcT>) frame, fgmask, modesUsed, if (detectShadows)
weight, variance, (PtrStepSz<WorkT>) mean, {
alphaT, alpha1, prune); cudaSafeCall(cudaFuncSetCacheConfig(mog2<true, SrcT, WorkT>, cudaFuncCachePreferL1));
}
cudaSafeCall( cudaGetLastError() ); mog2<true, SrcT, WorkT><<<grid, block, 0, stream>>>((PtrStepSz<SrcT>)frame, fgmask, modesUsed,
weight, variance, (PtrStepSz<WorkT>)mean,
alphaT, alpha1, prune, constants);
}
else
{
cudaSafeCall(cudaFuncSetCacheConfig(mog2<false, SrcT, WorkT>, cudaFuncCachePreferL1));
if (stream == 0) mog2<false, SrcT, WorkT><<<grid, block, 0, stream>>>((PtrStepSz<SrcT>)frame, fgmask, modesUsed,
cudaSafeCall( cudaDeviceSynchronize() ); weight, variance, (PtrStepSz<WorkT>)mean,
} alphaT, alpha1, prune, constants);
}
void mog2_gpu(PtrStepSzb frame, int cn, PtrStepSzb fgmask, PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzf variance, PtrStepSzb mean, cudaSafeCall(cudaGetLastError());
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, cudaStream_t stream);
static const func_t funcs[] = if (stream == 0)
{ cudaSafeCall(cudaDeviceSynchronize());
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); 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)
{
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);
template <typename WorkT, typename OutT> static const func_t funcs[] =
__global__ void getBackgroundImage2(const PtrStepSzb modesUsed, const PtrStepf gmm_weight, const PtrStep<WorkT> gmm_mean, PtrStep<OutT> dst)
{ {
const int x = blockIdx.x * blockDim.x + threadIdx.x; 0, mog2_caller<uchar, float>, 0, mog2_caller<uchar3, float3>, mog2_caller<uchar4, float4>};
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x >= modesUsed.cols || y >= modesUsed.rows) funcs[cn](frame, fgmask, modesUsed, weight, variance, mean, alphaT, prune, detectShadows, constants, stream);
return; }
int nmodes = modesUsed(y, x); template <typename WorkT, typename OutT>
__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;
WorkT meanVal = VecTraits<WorkT>::all(0.0f); if (x >= modesUsed.cols || y >= modesUsed.rows)
float totalWeight = 0.0f; return;
for (int mode = 0; mode < nmodes; ++mode) int nmodes = modesUsed(y, x);
{
float weight = gmm_weight(mode * modesUsed.rows + y, x);
WorkT mean = gmm_mean(mode * modesUsed.rows + y, x); WorkT meanVal = VecTraits<WorkT>::all(0.0f);
meanVal = meanVal + weight * mean; float totalWeight = 0.0f;
totalWeight += weight; for (int mode = 0; mode < nmodes; ++mode)
{
float weight = gmm_weight(mode * modesUsed.rows + y, x);
if(totalWeight > c_TB) WorkT mean = gmm_mean(mode * modesUsed.rows + y, x);
break; meanVal = meanVal + weight * mean;
}
meanVal = meanVal * (1.f / totalWeight); totalWeight += weight;
dst(y, x) = saturate_cast<OutT>(meanVal); if (totalWeight > constants->TB_)
} break;
}
template <typename WorkT, typename OutT> meanVal = meanVal * (1.f / totalWeight);
void getBackgroundImage2_caller(PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzb mean, PtrStepSzb dst, 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) ); dst(y, x) = saturate_cast<OutT>(meanVal);
}
getBackgroundImage2<WorkT, OutT><<<grid, block, 0, stream>>>(modesUsed, weight, (PtrStepSz<WorkT>) mean, (PtrStepSz<OutT>) dst); template <typename WorkT, typename OutT>
cudaSafeCall( cudaGetLastError() ); 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));
if (stream == 0) cudaSafeCall(cudaFuncSetCacheConfig(getBackgroundImage2<WorkT, OutT>, cudaFuncCachePreferL1));
cudaSafeCall( cudaDeviceSynchronize() );
}
void getBackgroundImage2_gpu(int cn, PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzb mean, PtrStepSzb dst, cudaStream_t stream) getBackgroundImage2<WorkT, OutT><<<grid, block, 0, stream>>>(modesUsed, weight, (PtrStepSz<WorkT>)mean, (PtrStepSz<OutT>)dst, constants);
{ cudaSafeCall(cudaGetLastError());
typedef void (*func_t)(PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzb mean, PtrStepSzb dst, cudaStream_t stream);
static const func_t funcs[] = if (stream == 0)
{ cudaSafeCall(cudaDeviceSynchronize());
0, getBackgroundImage2_caller<float, uchar>, 0, getBackgroundImage2_caller<float3, uchar3>, getBackgroundImage2_caller<float4, uchar4> }
};
funcs[cn](modesUsed, weight, mean, dst, 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, 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>};
funcs[cn](modesUsed, weight, mean, dst, constants, stream);
}
} // namespace mog2
} // namespace device
} // namespace cuda
} // namespace cv
#endif /* CUDA_DISABLER */ #endif /* CUDA_DISABLER */

@ -0,0 +1,37 @@
// 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 */

@ -41,209 +41,207 @@
//M*/ //M*/
#include "precomp.hpp" #include "precomp.hpp"
#include "cuda/mog2.hpp"
using namespace cv; using namespace cv;
using namespace cv::cuda; using namespace cv::cuda;
using namespace cv::cuda::device::mog2;
#if !defined HAVE_CUDA || defined(CUDA_DISABLER) #if !defined HAVE_CUDA || defined(CUDA_DISABLER)
Ptr<cuda::BackgroundSubtractorMOG2> cv::cuda::createBackgroundSubtractorMOG2(int, double, bool) { throw_no_cuda(); return Ptr<cuda::BackgroundSubtractorMOG2>(); } Ptr<cuda::BackgroundSubtractorMOG2> cv::cuda::createBackgroundSubtractorMOG2(int, double, bool)
{
throw_no_cuda();
return Ptr<cuda::BackgroundSubtractorMOG2>();
}
#else #else
namespace cv { namespace cuda { namespace device namespace
{ {
namespace mog2 // default parameters of gaussian background detection algorithm
{ const int defaultHistory = 500; // Learning rate; alpha = 1/defaultHistory2
void loadConstants(int nmixtures, float Tb, float TB, float Tg, float varInit, float varMin, float varMax, float tau, unsigned char shadowVal); const float defaultVarThreshold = 4.0f * 4.0f;
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); const int defaultNMixtures = 5; // maximal number of Gaussians in mixture
void getBackgroundImage2_gpu(int cn, PtrStepSzb modesUsed, PtrStepSzf weight, PtrStepSzb mean, PtrStepSzb dst, cudaStream_t stream); const float defaultBackgroundRatio = 0.9f; // threshold sum of weights for background test
} const float defaultVarThresholdGen = 3.0f * 3.0f;
}}} const float defaultVarInit = 15.0f; // initial variance for new components
const float defaultVarMax = 5.0f * defaultVarInit;
const float defaultVarMin = 4.0f;
// additional parameters
const float defaultCT = 0.05f; // complexity reduction prior constant 0 - no reduction of number of components
const unsigned char defaultShadowValue = 127; // value to use in the segmentation mask for shadows, set 0 not to do shadow detection
const float defaultShadowThreshold = 0.5f; // Tau - shadow threshold, see the paper for explanation
class MOG2Impl CV_FINAL : public cuda::BackgroundSubtractorMOG2
{
public:
MOG2Impl(int history, double varThreshold, bool detectShadows);
~MOG2Impl();
namespace void apply(InputArray image, OutputArray fgmask, double learningRate = -1) CV_OVERRIDE;
void apply(InputArray image, OutputArray fgmask, double learningRate, Stream &stream) CV_OVERRIDE;
void getBackgroundImage(OutputArray backgroundImage) const CV_OVERRIDE;
void getBackgroundImage(OutputArray backgroundImage, Stream &stream) const CV_OVERRIDE;
int getHistory() const CV_OVERRIDE { return history_; }
void setHistory(int history) CV_OVERRIDE { history_ = history; }
int getNMixtures() const CV_OVERRIDE { return constantsHost_.nmixtures_; }
void setNMixtures(int nmixtures) CV_OVERRIDE { constantsHost_.nmixtures_ = nmixtures; }
double getBackgroundRatio() const CV_OVERRIDE { return constantsHost_.TB_; }
void setBackgroundRatio(double ratio) CV_OVERRIDE { constantsHost_.TB_ = (float)ratio; }
double getVarThreshold() const CV_OVERRIDE { return constantsHost_.Tb_; }
void setVarThreshold(double varThreshold) CV_OVERRIDE { constantsHost_.Tb_ = (float)varThreshold; }
double getVarThresholdGen() const CV_OVERRIDE { return constantsHost_.Tg_; }
void setVarThresholdGen(double varThresholdGen) CV_OVERRIDE { constantsHost_.Tg_ = (float)varThresholdGen; }
double getVarInit() const CV_OVERRIDE { return constantsHost_.varInit_; }
void setVarInit(double varInit) CV_OVERRIDE { constantsHost_.varInit_ = (float)varInit; }
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 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; }
bool getDetectShadows() const CV_OVERRIDE { return detectShadows_; }
void setDetectShadows(bool detectShadows) CV_OVERRIDE { detectShadows_ = detectShadows; }
int getShadowValue() const CV_OVERRIDE { return constantsHost_.shadowVal_; }
void setShadowValue(int value) CV_OVERRIDE { constantsHost_.shadowVal_ = (uchar)value; }
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, Stream &stream);
Constants constantsHost_;
Constants *constantsDevice_;
int history_;
float ct_;
bool detectShadows_;
Size frameSize_;
int frameType_;
int nframes_;
GpuMat weight_;
GpuMat variance_;
GpuMat mean_;
//keep track of number of modes per pixel
GpuMat bgmodelUsedModes_;
};
MOG2Impl::MOG2Impl(int history, double varThreshold, bool detectShadows) : frameSize_(0, 0), frameType_(0), nframes_(0)
{
history_ = history > 0 ? history : defaultHistory;
detectShadows_ = detectShadows;
ct_ = defaultCT;
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)
{
apply(image, fgmask, learningRate, Stream::Null());
}
void MOG2Impl::apply(InputArray _frame, OutputArray _fgmask, double learningRate, Stream &stream)
{ {
// default parameters of gaussian background detection algorithm using namespace cv::cuda::device::mog2;
const int defaultHistory = 500; // Learning rate; alpha = 1/defaultHistory2
const float defaultVarThreshold = 4.0f * 4.0f; GpuMat frame = _frame.getGpuMat();
const int defaultNMixtures = 5; // maximal number of Gaussians in mixture
const float defaultBackgroundRatio = 0.9f; // threshold sum of weights for background test
const float defaultVarThresholdGen = 3.0f * 3.0f;
const float defaultVarInit = 15.0f; // initial variance for new components
const float defaultVarMax = 5.0f * defaultVarInit;
const float defaultVarMin = 4.0f;
// additional parameters
const float defaultCT = 0.05f; // complexity reduction prior constant 0 - no reduction of number of components
const unsigned char defaultShadowValue = 127; // value to use in the segmentation mask for shadows, set 0 not to do shadow detection
const float defaultShadowThreshold = 0.5f; // Tau - shadow threshold, see the paper for explanation
class MOG2Impl CV_FINAL : public cuda::BackgroundSubtractorMOG2
{
public:
MOG2Impl(int history, double varThreshold, bool detectShadows);
void apply(InputArray image, OutputArray fgmask, double learningRate=-1) CV_OVERRIDE;
void apply(InputArray image, OutputArray fgmask, double learningRate, Stream& stream) CV_OVERRIDE;
void getBackgroundImage(OutputArray backgroundImage) const CV_OVERRIDE;
void getBackgroundImage(OutputArray backgroundImage, Stream& stream) const CV_OVERRIDE;
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; }
double getBackgroundRatio() const CV_OVERRIDE { return backgroundRatio_; }
void setBackgroundRatio(double ratio) CV_OVERRIDE { backgroundRatio_ = (float) ratio; }
double getVarThreshold() const CV_OVERRIDE { return varThreshold_; }
void setVarThreshold(double varThreshold) CV_OVERRIDE { varThreshold_ = (float) varThreshold; }
double getVarThresholdGen() const CV_OVERRIDE { return varThresholdGen_; }
void setVarThresholdGen(double varThresholdGen) CV_OVERRIDE { varThresholdGen_ = (float) varThresholdGen; }
double getVarInit() const CV_OVERRIDE { return varInit_; }
void setVarInit(double varInit) CV_OVERRIDE { varInit_ = (float) varInit; }
double getVarMin() const CV_OVERRIDE { return varMin_; }
void setVarMin(double varMin) CV_OVERRIDE { varMin_ = (float) varMin; }
double getVarMax() const CV_OVERRIDE { return varMax_; }
void setVarMax(double varMax) CV_OVERRIDE { varMax_ = (float) varMax; }
double getComplexityReductionThreshold() const CV_OVERRIDE { return ct_; }
void setComplexityReductionThreshold(double ct) CV_OVERRIDE { ct_ = (float) ct; }
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; }
double getShadowThreshold() const CV_OVERRIDE { return shadowThreshold_; } int ch = frame.channels();
void setShadowThreshold(double threshold) CV_OVERRIDE { shadowThreshold_ = (float) threshold; } int work_ch = ch;
private:
void initialize(Size frameSize, int frameType);
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_;
int nframes_;
GpuMat weight_;
GpuMat variance_;
GpuMat mean_;
//keep track of number of modes per pixel
GpuMat bgmodelUsedModes_;
};
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;
}
void MOG2Impl::apply(InputArray image, OutputArray fgmask, double learningRate) if (nframes_ == 0 || learningRate >= 1.0 || frame.size() != frameSize_ || work_ch != mean_.channels())
{ initialize(frame.size(), frame.type(), stream);
apply(image, fgmask, learningRate, Stream::Null());
}
void MOG2Impl::apply(InputArray _frame, OutputArray _fgmask, double learningRate, Stream& stream) _fgmask.create(frameSize_, CV_8UC1);
{ GpuMat fgmask = _fgmask.getGpuMat();
using namespace cv::cuda::device::mog2;
GpuMat frame = _frame.getGpuMat(); fgmask.setTo(Scalar::all(0), stream);
int ch = frame.channels(); ++nframes_;
int work_ch = ch; learningRate = learningRate >= 0 && nframes_ > 1 ? learningRate : 1.0 / std::min(2 * nframes_, history_);
CV_Assert(learningRate >= 0);
if (nframes_ == 0 || learningRate >= 1.0 || frame.size() != frameSize_ || work_ch != mean_.channels()) mog2_gpu(frame, frame.channels(), fgmask, bgmodelUsedModes_, weight_, variance_, mean_,
initialize(frame.size(), frame.type()); (float)learningRate, static_cast<float>(-learningRate * ct_), detectShadows_, constantsDevice_, StreamAccessor::getStream(stream));
}
void MOG2Impl::getBackgroundImage(OutputArray backgroundImage) const
{
getBackgroundImage(backgroundImage, Stream::Null());
}
void MOG2Impl::getBackgroundImage(OutputArray _backgroundImage, Stream &stream) const
{
using namespace cv::cuda::device::mog2;
_backgroundImage.create(frameSize_, frameType_);
GpuMat backgroundImage = _backgroundImage.getGpuMat();
getBackgroundImage2_gpu(backgroundImage.channels(), bgmodelUsedModes_, weight_, mean_, backgroundImage, constantsDevice_, StreamAccessor::getStream(stream));
}
void MOG2Impl::initialize(cv::Size frameSize, int frameType, Stream &stream)
{
using namespace cv::cuda::device::mog2;
_fgmask.create(frameSize_, CV_8UC1); CV_Assert(frameType == CV_8UC1 || frameType == CV_8UC3 || frameType == CV_8UC4);
GpuMat fgmask = _fgmask.getGpuMat();
fgmask.setTo(Scalar::all(0), stream); frameSize_ = frameSize;
frameType_ = frameType;
nframes_ = 0;
++nframes_; const int ch = CV_MAT_CN(frameType);
learningRate = learningRate >= 0 && nframes_ > 1 ? learningRate : 1.0 / std::min(2 * nframes_, history_); const int work_ch = ch;
CV_Assert( learningRate >= 0 );
mog2_gpu(frame, frame.channels(), fgmask, bgmodelUsedModes_, weight_, variance_, mean_, // for each gaussian mixture of each pixel bg model we store ...
(float) learningRate, static_cast<float>(-learningRate * ct_), detectShadows_, StreamAccessor::getStream(stream)); // the mixture weight (w),
} // the mean (nchannels values) and
// the covariance
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));
void MOG2Impl::getBackgroundImage(OutputArray backgroundImage) const //make the array for keeping track of the used modes per pixel - all zeros at start
{ bgmodelUsedModes_.create(frameSize_, CV_8UC1);
getBackgroundImage(backgroundImage, Stream::Null()); bgmodelUsedModes_.setTo(Scalar::all(0));
}
void MOG2Impl::getBackgroundImage(OutputArray _backgroundImage, Stream& stream) const
{
using namespace cv::cuda::device::mog2;
_backgroundImage.create(frameSize_, frameType_);
GpuMat backgroundImage = _backgroundImage.getGpuMat();
getBackgroundImage2_gpu(backgroundImage.channels(), bgmodelUsedModes_, weight_, mean_, backgroundImage, StreamAccessor::getStream(stream));
}
void MOG2Impl::initialize(cv::Size frameSize, int frameType)
{
using namespace cv::cuda::device::mog2;
CV_Assert( frameType == CV_8UC1 || frameType == CV_8UC3 || frameType == CV_8UC4 );
frameSize_ = frameSize;
frameType_ = frameType;
nframes_ = 0;
int ch = CV_MAT_CN(frameType);
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));
//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) Ptr<cuda::BackgroundSubtractorMOG2> cv::cuda::createBackgroundSubtractorMOG2(int history, double varThreshold, bool detectShadows)
{ {

Loading…
Cancel
Save