};
+//! Calculates optical flow for 2 images using block matching algorithm */
+CV_EXPORTS void calcOpticalFlowBM(const GpuMat& prev, const GpuMat& curr,
+ Size block_size, Size shift_size, Size max_range, bool use_previous,
+ GpuMat& velx, GpuMat& vely, GpuMat& buf,
+ Stream& stream = Stream::Null());
+
+
//! Interpolate frames (images) using provided optical flow (displacement field).
//! frame0 - frame 0 (32-bit floating point images, single channel)
//! frame1 - frame 1 (the same type and size)
}
//////////////////////////////////////////////////////
+// OpticalFlowBM
+
+void calcOpticalFlowBM(const cv::Mat& prev, const cv::Mat& curr,
+ cv::Size bSize, cv::Size shiftSize, cv::Size maxRange, int usePrevious,
+ cv::Mat& velx, cv::Mat& vely)
+{
+ cv::Size sz((curr.cols - bSize.width + shiftSize.width)/shiftSize.width, (curr.rows - bSize.height + shiftSize.height)/shiftSize.height);
+
+ velx.create(sz, CV_32FC1);
+ vely.create(sz, CV_32FC1);
+
+ CvMat cvprev = prev;
+ CvMat cvcurr = curr;
+
+ CvMat cvvelx = velx;
+ CvMat cvvely = vely;
+
+ cvCalcOpticalFlowBM(&cvprev, &cvcurr, bSize, shiftSize, maxRange, usePrevious, &cvvelx, &cvvely);
+}
+
+PERF_TEST_P(ImagePair, Video_OpticalFlowBM,
+ Values<pair_string>(make_pair("gpu/opticalflow/frame0.png", "gpu/opticalflow/frame1.png")))
+{
+ declare.time(400);
+
+ cv::Mat frame0 = readImage(GetParam().first, cv::IMREAD_GRAYSCALE);
+ ASSERT_FALSE(frame0.empty());
+
+ cv::Mat frame1 = readImage(GetParam().second, cv::IMREAD_GRAYSCALE);
+ ASSERT_FALSE(frame1.empty());
+
+ cv::Size block_size(16, 16);
+ cv::Size shift_size(1, 1);
+ cv::Size max_range(16, 16);
+
+ if (PERF_RUN_GPU())
+ {
+ cv::gpu::GpuMat d_frame0(frame0);
+ cv::gpu::GpuMat d_frame1(frame1);
+ cv::gpu::GpuMat d_velx, d_vely, buf;
+
+ cv::gpu::calcOpticalFlowBM(d_frame0, d_frame1, block_size, shift_size, max_range, false, d_velx, d_vely, buf);
+
+ TEST_CYCLE()
+ {
+ cv::gpu::calcOpticalFlowBM(d_frame0, d_frame1, block_size, shift_size, max_range, false, d_velx, d_vely, buf);
+ }
+
+ GPU_SANITY_CHECK(d_velx);
+ GPU_SANITY_CHECK(d_vely);
+ }
+ else
+ {
+ cv::Mat velx, vely;
+
+ calcOpticalFlowBM(frame0, frame1, block_size, shift_size, max_range, false, velx, vely);
+
+ TEST_CYCLE()
+ {
+ calcOpticalFlowBM(frame0, frame1, block_size, shift_size, max_range, false, velx, vely);
+ }
+
+ CPU_SANITY_CHECK(velx);
+ CPU_SANITY_CHECK(vely);
+ }
+}
+
+//////////////////////////////////////////////////////
// FGDStatModel
DEF_PARAM_TEST_1(Video, string);
--- /dev/null
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+// By downloading, copying, installing or using the software you agree to this license.
+// If you do not agree to this license, do not download, install,
+// copy or use the software.
+//
+//
+// License Agreement
+// For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+// * Redistribution's of source code must retain the above copyright notice,
+// this list of conditions and the following disclaimer.
+//
+// * Redistribution's in binary form must reproduce the above copyright notice,
+// this list of conditions and the following disclaimer in the documentation
+// and/or other materials provided with the distribution.
+//
+// * The name of the copyright holders may not be used to endorse or promote products
+// derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or bpied warranties, including, but not limited to, the bpied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#if !defined CUDA_DISABLER
+
+#include "opencv2/gpu/device/common.hpp"
+#include "opencv2/gpu/device/limits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::device;
+
+namespace
+{
+ texture<uchar, cudaTextureType2D, cudaReadModeElementType> tex_prev(false, cudaFilterModePoint, cudaAddressModeClamp);
+ texture<uchar, cudaTextureType2D, cudaReadModeElementType> tex_curr(false, cudaFilterModePoint, cudaAddressModeClamp);
+
+ __device__ int cmpBlocks(int X1, int Y1, int X2, int Y2, int2 blockSize)
+ {
+ int s = 0;
+
+ for (int y = 0; y < blockSize.y; ++y)
+ {
+ for (int x = 0; x < blockSize.x; ++x)
+ s += ::abs(tex2D(tex_prev, X1 + x, Y1 + y) - tex2D(tex_curr, X2 + x, Y2 + y));
+ }
+
+ return s;
+ }
+
+ __global__ void calcOptFlowBM(PtrStepSzf velx, PtrStepf vely, const int2 blockSize, const int2 shiftSize, const bool usePrevious,
+ const int maxX, const int maxY, const int acceptLevel, const int escapeLevel,
+ const short2* ss, const int ssCount)
+ {
+ const int j = blockIdx.x * blockDim.x + threadIdx.x;
+ const int i = blockIdx.y * blockDim.y + threadIdx.y;
+
+ if (i >= velx.rows || j >= velx.cols)
+ return;
+
+ const int X1 = j * shiftSize.x;
+ const int Y1 = i * shiftSize.y;
+
+ const int offX = usePrevious ? __float2int_rn(velx(i, j)) : 0;
+ const int offY = usePrevious ? __float2int_rn(vely(i, j)) : 0;
+
+ int X2 = X1 + offX;
+ int Y2 = Y1 + offY;
+
+ int dist = numeric_limits<int>::max();
+
+ if (0 <= X2 && X2 <= maxX && 0 <= Y2 && Y2 <= maxY)
+ dist = cmpBlocks(X1, Y1, X2, Y2, blockSize);
+
+ int countMin = 1;
+ int sumx = offX;
+ int sumy = offY;
+
+ if (dist > acceptLevel)
+ {
+ // do brute-force search
+ for (int k = 0; k < ssCount; ++k)
+ {
+ const short2 ssVal = ss[k];
+
+ const int dx = offX + ssVal.x;
+ const int dy = offY + ssVal.y;
+
+ X2 = X1 + dx;
+ Y2 = Y1 + dy;
+
+ if (0 <= X2 && X2 <= maxX && 0 <= Y2 && Y2 <= maxY)
+ {
+ const int tmpDist = cmpBlocks(X1, Y1, X2, Y2, blockSize);
+ if (tmpDist < acceptLevel)
+ {
+ sumx = dx;
+ sumy = dy;
+ countMin = 1;
+ break;
+ }
+
+ if (tmpDist < dist)
+ {
+ dist = tmpDist;
+ sumx = dx;
+ sumy = dy;
+ countMin = 1;
+ }
+ else if (tmpDist == dist)
+ {
+ sumx += dx;
+ sumy += dy;
+ countMin++;
+ }
+ }
+ }
+
+ if (dist > escapeLevel)
+ {
+ sumx = offX;
+ sumy = offY;
+ countMin = 1;
+ }
+ }
+
+ velx(i, j) = static_cast<float>(sumx) / countMin;
+ vely(i, j) = static_cast<float>(sumy) / countMin;
+ }
+}
+
+namespace optflowbm
+{
+ void calc(PtrStepSzb prev, PtrStepSzb curr, PtrStepSzf velx, PtrStepSzf vely, int2 blockSize, int2 shiftSize, bool usePrevious,
+ int maxX, int maxY, int acceptLevel, int escapeLevel, const short2* ss, int ssCount, cudaStream_t stream)
+ {
+ bindTexture(&tex_prev, prev);
+ bindTexture(&tex_curr, curr);
+
+ const dim3 block(32, 8);
+ const dim3 grid(divUp(velx.cols, block.x), divUp(vely.rows, block.y));
+
+ calcOptFlowBM<<<grid, block, 0, stream>>>(velx, vely, blockSize, shiftSize, usePrevious,
+ maxX, maxY, acceptLevel, escapeLevel, ss, ssCount);
+ cudaSafeCall( cudaGetLastError() );
+
+ if (stream == 0)
+ cudaSafeCall( cudaDeviceSynchronize() );
+ }
+}
+
+#endif // !defined CUDA_DISABLER
--- /dev/null
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+// By downloading, copying, installing or using the software you agree to this license.
+// If you do not agree to this license, do not download, install,
+// copy or use the software.
+//
+//
+// License Agreement
+// For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+// * Redistribution's of source code must retain the above copyright notice,
+// this list of conditions and the following disclaimer.
+//
+// * Redistribution's in binary form must reproduce the above copyright notice,
+// this list of conditions and the following disclaimer in the documentation
+// and/or other materials provided with the distribution.
+//
+// * The name of the copyright holders may not be used to endorse or promote products
+// derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "precomp.hpp"
+
+using namespace std;
+using namespace cv;
+using namespace cv::gpu;
+
+#if !defined HAVE_CUDA || defined(CUDA_DISABLER)
+
+void cv::gpu::calcOpticalFlowBM(const GpuMat&, const GpuMat&, Size, Size, Size, bool, GpuMat&, GpuMat&, GpuMat&, Stream&) { throw_nogpu(); }
+
+#else // HAVE_CUDA
+
+namespace optflowbm
+{
+ void calc(PtrStepSzb prev, PtrStepSzb curr, PtrStepSzf velx, PtrStepSzf vely, int2 blockSize, int2 shiftSize, bool usePrevious,
+ int maxX, int maxY, int acceptLevel, int escapeLevel, const short2* ss, int ssCount, cudaStream_t stream);
+}
+
+void cv::gpu::calcOpticalFlowBM(const GpuMat& prev, const GpuMat& curr, Size blockSize, Size shiftSize, Size maxRange, bool usePrevious, GpuMat& velx, GpuMat& vely, GpuMat& buf, Stream& st)
+{
+ CV_Assert( prev.type() == CV_8UC1 );
+ CV_Assert( curr.size() == prev.size() && curr.type() == prev.type() );
+
+ const Size velSize((prev.cols - blockSize.width + shiftSize.width) / shiftSize.width,
+ (prev.rows - blockSize.height + shiftSize.height) / shiftSize.height);
+
+ velx.create(velSize, CV_32FC1);
+ vely.create(velSize, CV_32FC1);
+
+ // scanning scheme coordinates
+ vector<short2> ss((2 * maxRange.width + 1) * (2 * maxRange.height + 1));
+ int ssCount = 0;
+
+ // Calculate scanning scheme
+ const int minCount = std::min(maxRange.width, maxRange.height);
+
+ // use spiral search pattern
+ //
+ // 9 10 11 12
+ // 8 1 2 13
+ // 7 * 3 14
+ // 6 5 4 15
+ //... 20 19 18 17
+ //
+
+ for (int i = 0; i < minCount; ++i)
+ {
+ // four cycles along sides
+ int x = -i - 1, y = x;
+
+ // upper side
+ for (int j = -i; j <= i + 1; ++j, ++ssCount)
+ {
+ ss[ssCount].x = ++x;
+ ss[ssCount].y = y;
+ }
+
+ // right side
+ for (int j = -i; j <= i + 1; ++j, ++ssCount)
+ {
+ ss[ssCount].x = x;
+ ss[ssCount].y = ++y;
+ }
+
+ // bottom side
+ for (int j = -i; j <= i + 1; ++j, ++ssCount)
+ {
+ ss[ssCount].x = --x;
+ ss[ssCount].y = y;
+ }
+
+ // left side
+ for (int j = -i; j <= i + 1; ++j, ++ssCount)
+ {
+ ss[ssCount].x = x;
+ ss[ssCount].y = --y;
+ }
+ }
+
+ // the rest part
+ if (maxRange.width < maxRange.height)
+ {
+ const int xleft = -minCount;
+
+ // cycle by neighbor rings
+ for (int i = minCount; i < maxRange.height; ++i)
+ {
+ // two cycles by x
+ int y = -(i + 1);
+ int x = xleft;
+
+ // upper side
+ for (int j = -maxRange.width; j <= maxRange.width; ++j, ++ssCount, ++x)
+ {
+ ss[ssCount].x = x;
+ ss[ssCount].y = y;
+ }
+
+ x = xleft;
+ y = -y;
+
+ // bottom side
+ for (int j = -maxRange.width; j <= maxRange.width; ++j, ++ssCount, ++x)
+ {
+ ss[ssCount].x = x;
+ ss[ssCount].y = y;
+ }
+ }
+ }
+ else if (maxRange.width > maxRange.height)
+ {
+ const int yupper = -minCount;
+
+ // cycle by neighbor rings
+ for (int i = minCount; i < maxRange.width; ++i)
+ {
+ // two cycles by y
+ int x = -(i + 1);
+ int y = yupper;
+
+ // left side
+ for (int j = -maxRange.height; j <= maxRange.height; ++j, ++ssCount, ++y)
+ {
+ ss[ssCount].x = x;
+ ss[ssCount].y = y;
+ }
+
+ y = yupper;
+ x = -x;
+
+ // right side
+ for (int j = -maxRange.height; j <= maxRange.height; ++j, ++ssCount, ++y)
+ {
+ ss[ssCount].x = x;
+ ss[ssCount].y = y;
+ }
+ }
+ }
+
+ const cudaStream_t stream = StreamAccessor::getStream(st);
+
+ ensureSizeIsEnough(1, ssCount, CV_16SC2, buf);
+ if (stream == 0)
+ cudaSafeCall( cudaMemcpy(buf.data, &ss[0], ssCount * sizeof(short2), cudaMemcpyHostToDevice) );
+ else
+ cudaSafeCall( cudaMemcpyAsync(buf.data, &ss[0], ssCount * sizeof(short2), cudaMemcpyHostToDevice, stream) );
+
+ const int maxX = prev.cols - blockSize.width;
+ const int maxY = prev.rows - blockSize.height;
+
+ const int SMALL_DIFF = 2;
+ const int BIG_DIFF = 128;
+
+ const int blSize = blockSize.area();
+ const int acceptLevel = blSize * SMALL_DIFF;
+ const int escapeLevel = blSize * BIG_DIFF;
+
+ optflowbm::calc(prev, curr, velx, vely,
+ make_int2(blockSize.width, blockSize.height), make_int2(shiftSize.width, shiftSize.height), usePrevious,
+ maxX, maxY, acceptLevel, escapeLevel, buf.ptr<short2>(), ssCount, stream);
+}
+
+#endif // HAVE_CUDA
WHOLE_SUBMAT));
//////////////////////////////////////////////////////
+// OpticalFlowBM
+
+void calcOpticalFlowBM(const cv::Mat& prev, const cv::Mat& curr,
+ cv::Size bSize, cv::Size shiftSize, cv::Size maxRange, int usePrevious,
+ cv::Mat& velx, cv::Mat& vely)
+{
+ cv::Size sz((curr.cols - bSize.width + shiftSize.width)/shiftSize.width, (curr.rows - bSize.height + shiftSize.height)/shiftSize.height);
+
+ velx.create(sz, CV_32FC1);
+ vely.create(sz, CV_32FC1);
+
+ CvMat cvprev = prev;
+ CvMat cvcurr = curr;
+
+ CvMat cvvelx = velx;
+ CvMat cvvely = vely;
+
+ cvCalcOpticalFlowBM(&cvprev, &cvcurr, bSize, shiftSize, maxRange, usePrevious, &cvvelx, &cvvely);
+}
+
+struct OpticalFlowBM : testing::TestWithParam<cv::gpu::DeviceInfo>
+{
+};
+
+TEST_P(OpticalFlowBM, Accuracy)
+{
+ cv::gpu::DeviceInfo devInfo = GetParam();
+ cv::gpu::setDevice(devInfo.deviceID());
+
+ cv::Mat frame0 = readImage("opticalflow/rubberwhale1.png", cv::IMREAD_GRAYSCALE);
+ ASSERT_FALSE(frame0.empty());
+
+ cv::Mat frame1 = readImage("opticalflow/rubberwhale2.png", cv::IMREAD_GRAYSCALE);
+ ASSERT_FALSE(frame1.empty());
+
+ cv::Size block_size(16, 16);
+ cv::Size shift_size(1, 1);
+ cv::Size max_range(16, 16);
+
+ cv::gpu::GpuMat d_velx, d_vely, buf;
+ cv::gpu::calcOpticalFlowBM(loadMat(frame0), loadMat(frame1),
+ block_size, shift_size, max_range, false,
+ d_velx, d_vely, buf);
+
+ cv::Mat velx, vely;
+ calcOpticalFlowBM(frame0, frame1, block_size, shift_size, max_range, false, velx, vely);
+
+ EXPECT_MAT_NEAR(velx, d_velx, 0);
+ EXPECT_MAT_NEAR(vely, d_vely, 0);
+}
+
+INSTANTIATE_TEST_CASE_P(GPU_Video, OpticalFlowBM, ALL_DEVICES);
+
+//////////////////////////////////////////////////////
// FGDStatModel
namespace cv