generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

added support of CV_TM_CCORR (via FFT) into gpu::matchTemplate (versions both with block and without blocks)

Browse Source
This commit is contained in:
Alexey Spizhevoy
2010-12-08 13:03:53 +00:00
parent 3beac049d5
commit 40304721a7
4 changed files with 301 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
modules/gpu/src/cuda/match_template.cu
View File

@@ -40,8 +40,12 @@
//
//M*/
#include <cufft.h>
#include "internal_shared.hpp"
#include <iostream>
using namespace std;
using namespace cv::gpu;
namespace cv { namespace gpu { namespace imgproc {
@@ -50,7 +54,7 @@ texture<unsigned char, 2> imageTex_8U;
texture<unsigned char, 2> templTex_8U;
__global__ void matchTemplateKernel_8U_SqDiff(int w, int h, DevMem2Df result)
__global__ void matchTemplateKernel_8U_SQDIFF(int w, int h, DevMem2Df result)
{
int x = blockDim.x * blockIdx.x + threadIdx.x;
int y = blockDim.y * blockIdx.y + threadIdx.y;
@@ -75,7 +79,7 @@ __global__ void matchTemplateKernel_8U_SqDiff(int w, int h, DevMem2Df result)
}
void matchTemplateCaller_8U_SqDiff(const DevMem2D image, const DevMem2D templ, DevMem2Df result)
void matchTemplate_8U_SQDIFF(const DevMem2D image, const DevMem2D templ, DevMem2Df result)
{
dim3 threads(32, 8);
dim3 grid(divUp(image.cols - templ.cols + 1, threads.x),
@@ -87,11 +91,32 @@ void matchTemplateCaller_8U_SqDiff(const DevMem2D image, const DevMem2D templ, D
imageTex_8U.filterMode = cudaFilterModePoint;
templTex_8U.filterMode = cudaFilterModePoint;
matchTemplateKernel_8U_SqDiff<<<grid, threads>>>(templ.cols, templ.rows, result);
matchTemplateKernel_8U_SQDIFF<<<grid, threads>>>(templ.cols, templ.rows, result);
cudaSafeCall(cudaThreadSynchronize());
cudaSafeCall(cudaUnbindTexture(imageTex_8U));
cudaSafeCall(cudaUnbindTexture(templTex_8U));
}
__global__ void multiplyAndNormalizeSpectsKernel(int n, float scale, const cufftComplex* a,
const cufftComplex* b, cufftComplex* c)
{
int x = blockIdx.x * blockDim.x + threadIdx.x;
if (x < n)
{
cufftComplex v = cuCmulf(a[x], cuConjf(b[x]));
c[x] = make_cuFloatComplex(cuCrealf(v) * scale, cuCimagf(v) * scale);
}
}
void multiplyAndNormalizeSpects(int n, float scale, const cufftComplex* a, const cufftComplex* b,
cufftComplex* c)
{
dim3 threads(256);
dim3 grid(divUp(n, threads.x));
multiplyAndNormalizeSpectsKernel<<<grid, threads>>>(n, scale, a, b, c);
}
}}}