Chiebot-Mirror
/
opencv
mirror of https://github.com/opencv/opencv.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

refactored generalized hough (both CPU and GPU):

Browse Source 
removed set/get methods from Algorithm (implement owns)
removed GHT_* enumeration
pull/1149/head
Vladislav Vinogradov 12 years ago
parent 4b234fa0a5
commit 7b87d72d80
9 changed files with 1028 additions and 1935 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 93
      modules/gpuimgproc/doc/hough.rst
  2. 21
      modules/gpuimgproc/include/opencv2/gpuimgproc.hpp
  3. 91
      modules/gpuimgproc/perf/perf_hough.cpp
  4. 262
      modules/gpuimgproc/src/cuda/generalized_hough.cu
  5. 1113
      modules/gpuimgproc/src/generalized_hough.cpp
  6. 12
      modules/gpuimgproc/test/test_hough.cpp
  7. 119
      modules/imgproc/include/opencv2/imgproc.hpp
  8. 1121
      modules/imgproc/src/generalized_hough.cpp
  9. 131
      samples/gpu/generalized_hough.cpp

93
modules/gpuimgproc/doc/hough.rst
Unescape View File

@ -213,98 +213,19 @@ Creates implementation for :ocv:class:`gpu::HoughCirclesDetector` .
gpu::GeneralizedHough
---------------------
.. ocv:class:: gpu::GeneralizedHough : public Algorithm
Base class for generalized hough transform. ::
class CV_EXPORTS GeneralizedHough : public Algorithm
{
public:
static Ptr<GeneralizedHough> create(int method);
virtual void setTemplate(InputArray templ, int cannyThreshold = 100, Point templCenter = Point(-1, -1)) = 0;
virtual void setTemplate(InputArray edges, InputArray dx, InputArray dy, Point templCenter = Point(-1, -1)) = 0;
virtual void detect(InputArray image, OutputArray positions, int cannyThreshold = 100) = 0;
virtual void detect(InputArray edges, InputArray dx, InputArray dy, OutputArray positions) = 0;
virtual void downloadResults(InputArray d_positions, OutputArray h_positions, OutputArray h_votes = noArray()) = 0;
};
Finds arbitrary template in the grayscale image using Generalized Hough Transform.
gpu::GeneralizedHough::create
-----------------------------
Creates implementation for :ocv:class:`gpu::GeneralizedHough` .
.. ocv:function:: Ptr<GeneralizedHough> gpu::GeneralizedHough::create(int method)
:param method: Combination of flags ( ``cv::GeneralizedHough::GHT_POSITION`` , ``cv::GeneralizedHough::GHT_SCALE`` , ``cv::GeneralizedHough::GHT_ROTATION`` ) specifying transformation to find.
For full affine transformations (move + scale + rotation) [Guil1999]_ algorithm is used, otherwise [Ballard1981]_ algorithm is used.
gpu::GeneralizedHough::setTemplate
gpu::createGeneralizedHoughBallard
----------------------------------
Set template to search.
.. ocv:function:: void gpu::GeneralizedHough::setTemplate(InputArray templ, int cannyThreshold = 100, Point templCenter = Point(-1, -1))
.. ocv:function:: void gpu::GeneralizedHough::setTemplate(InputArray edges, InputArray dx, InputArray dy, Point templCenter = Point(-1, -1))
Creates implementation for generalized hough transform from [Ballard1981]_ .
:param templ: Template image. Canny edge detector will be applied to extract template edges.
.. ocv:function:: Ptr<GeneralizedHoughBallard> gpu::createGeneralizedHoughBallard()
:param cannyThreshold: Threshold value for Canny edge detector.
:param templCenter: Center for rotation. By default image center will be used.
:param edges: Edge map for template image.
:param dx: First derivative of template image in the vertical direction. Support only ``CV_32S`` type.
:param dy: First derivative of template image in the horizontal direction. Support only ``CV_32S`` type.
gpu::GeneralizedHough::detect
-----------------------------
Finds template (set by :ocv:func:`gpu::GeneralizedHough::setTemplate` ) in the grayscale image.
.. ocv:function:: void gpu::GeneralizedHough::detect(InputArray image, OutputArray positions, int cannyThreshold = 100)
.. ocv:function:: void gpu::GeneralizedHough::detect(InputArray edges, InputArray dx, InputArray dy, OutputArray positions)
:param templ: Input image. Canny edge detector will be applied to extract template edges.
:param positions: Output vector of found objects. Each vector is encoded as a 4-element floating-point vector :math:`(x, y, scale, angle)` .
:param cannyThreshold: Threshold value for Canny edge detector.
:param edges: Edge map for input image.
:param dx: First derivative of input image in the vertical direction. Support only ``CV_32S`` type.
:param dy: First derivative of input image in the horizontal direction. Support only ``CV_32S`` type.
gpu::GeneralizedHough::downloadResults
--------------------------------------
Downloads results from :ocv:func:`gpu::GeneralizedHough::detect` to host memory.
.. ocv:function:: void gpu::GeneralizedHough::downloadResult(InputArray d_positions, OutputArray h_positions, OutputArray h_votes = noArray())
:param d_lines: Result of :ocv:func:`gpu::GeneralizedHough::detect` .
:param h_lines: Output host array.
gpu::createGeneralizedHoughGuil
-------------------------------
Creates implementation for generalized hough transform from [Guil1999]_ .
:param h_votes: Optional output array for votes. Each vector is encoded as a 3-element integer-point vector :math:`(position_votes, scale_votes, angle_votes)` .
.. ocv:function:: Ptr<GeneralizedHoughGuil> gpu::createGeneralizedHoughGuil()

21
modules/gpuimgproc/include/opencv2/gpuimgproc.hpp
Unescape View File

@ -283,24 +283,13 @@ CV_EXPORTS Ptr<HoughCirclesDetector> createHoughCirclesDetector(float dp, float
//////////////////////////////////////
// GeneralizedHough
//! finds arbitrary template in the grayscale image using Generalized Hough Transform
//! Ballard, D.H. (1981). Generalizing the Hough transform to detect arbitrary shapes. Pattern Recognition 13 (2): 111-122.
//! Guil, N., González-Linares, J.M. and Zapata, E.L. (1999). Bidimensional shape detection using an invariant approach. Pattern Recognition 32 (6): 1025-1038.
class CV_EXPORTS GeneralizedHough : public Algorithm
{
public:
static Ptr<GeneralizedHough> create(int method);
//! set template to search
virtual void setTemplate(InputArray templ, int cannyThreshold = 100, Point templCenter = Point(-1, -1)) = 0;
virtual void setTemplate(InputArray edges, InputArray dx, InputArray dy, Point templCenter = Point(-1, -1)) = 0;
//! Detects position only without traslation and rotation
CV_EXPORTS Ptr<GeneralizedHoughBallard> createGeneralizedHoughBallard();
//! find template on image
virtual void detect(InputArray image, OutputArray positions, int cannyThreshold = 100) = 0;
virtual void detect(InputArray edges, InputArray dx, InputArray dy, OutputArray positions) = 0;
virtual void downloadResults(InputArray d_positions, OutputArray h_positions, OutputArray h_votes = noArray()) = 0;
};
//! Guil, N., González-Linares, J.M. and Zapata, E.L. (1999). Bidimensional shape detection using an invariant approach. Pattern Recognition 32 (6): 1025-1038.
//! Detects position, traslation and rotation
CV_EXPORTS Ptr<GeneralizedHoughGuil> createGeneralizedHoughGuil();
////////////////////////// Corners Detection ///////////////////////////

91
modules/gpuimgproc/perf/perf_hough.cpp
Unescape View File

@ -227,23 +227,59 @@ PERF_TEST_P(Sz_Dp_MinDist, HoughCircles,
//////////////////////////////////////////////////////////////////////
// GeneralizedHough
enum { GHT_POSITION = cv::GeneralizedHough::GHT_POSITION,
GHT_SCALE = cv::GeneralizedHough::GHT_SCALE,
GHT_ROTATION = cv::GeneralizedHough::GHT_ROTATION
};
PERF_TEST_P(Sz, GeneralizedHoughBallard, GPU_TYPICAL_MAT_SIZES)
{
declare.time(10);
const cv::Size imageSize = GetParam();
const cv::Mat templ = readImage("cv/shared/templ.png", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(templ.empty());
cv::Mat image(imageSize, CV_8UC1, cv::Scalar::all(0));
templ.copyTo(image(cv::Rect(50, 50, templ.cols, templ.rows)));
cv::Mat edges;
cv::Canny(image, edges, 50, 100);
cv::Mat dx, dy;
cv::Sobel(image, dx, CV_32F, 1, 0);
cv::Sobel(image, dy, CV_32F, 0, 1);
if (PERF_RUN_GPU())
{
cv::Ptr<cv::GeneralizedHoughBallard> alg = cv::gpu::createGeneralizedHoughBallard();
const cv::gpu::GpuMat d_edges(edges);
const cv::gpu::GpuMat d_dx(dx);
const cv::gpu::GpuMat d_dy(dy);
cv::gpu::GpuMat positions;
alg->setTemplate(cv::gpu::GpuMat(templ));
TEST_CYCLE() alg->detect(d_edges, d_dx, d_dy, positions);
GPU_SANITY_CHECK(positions);
}
else
{
cv::Ptr<cv::GeneralizedHoughBallard> alg = cv::createGeneralizedHoughBallard();
cv::Mat positions;
alg->setTemplate(templ);
CV_FLAGS(GHMethod, GHT_POSITION, GHT_SCALE, GHT_ROTATION);
TEST_CYCLE() alg->detect(edges, dx, dy, positions);
DEF_PARAM_TEST(Method_Sz, GHMethod, cv::Size);
CPU_SANITY_CHECK(positions);
}
}
PERF_TEST_P(Method_Sz, GeneralizedHough,
Combine(Values(GHMethod(GHT_POSITION), GHMethod(GHT_POSITION | GHT_SCALE), GHMethod(GHT_POSITION | GHT_ROTATION), GHMethod(GHT_POSITION | GHT_SCALE | GHT_ROTATION)),
GPU_TYPICAL_MAT_SIZES))
PERF_TEST_P(Sz, GeneralizedHoughGuil, GPU_TYPICAL_MAT_SIZES)
{
declare.time(10);
const int method = GET_PARAM(0);
const cv::Size imageSize = GET_PARAM(1);
const cv::Size imageSize = GetParam();
const cv::Mat templ = readImage("cv/shared/templ.png", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(templ.empty());
@ -281,39 +317,32 @@ PERF_TEST_P(Method_Sz, GeneralizedHough,
if (PERF_RUN_GPU())
{
cv::Ptr<cv::GeneralizedHoughGuil> alg = cv::gpu::createGeneralizedHoughGuil();
alg->setMaxAngle(90.0);
alg->setAngleStep(2.0);
const cv::gpu::GpuMat d_edges(edges);
const cv::gpu::GpuMat d_dx(dx);
const cv::gpu::GpuMat d_dy(dy);
cv::gpu::GpuMat posAndVotes;
cv::gpu::GpuMat positions;
cv::Ptr<cv::gpu::GeneralizedHough> d_hough = cv::gpu::GeneralizedHough::create(method);
if (method & GHT_ROTATION)
{
d_hough->set("maxAngle", 90.0);
d_hough->set("angleStep", 2.0);
}
alg->setTemplate(cv::gpu::GpuMat(templ));
d_hough->setTemplate(cv::gpu::GpuMat(templ));
TEST_CYCLE() alg->detect(d_edges, d_dx, d_dy, positions);
TEST_CYCLE() d_hough->detect(d_edges, d_dx, d_dy, posAndVotes);
const cv::gpu::GpuMat positions(1, posAndVotes.cols, CV_32FC4, posAndVotes.data);
GPU_SANITY_CHECK(positions);
}
else
{
cv::Mat positions;
cv::Ptr<cv::GeneralizedHoughGuil> alg = cv::createGeneralizedHoughGuil();
alg->setMaxAngle(90.0);
alg->setAngleStep(2.0);
cv::Ptr<cv::GeneralizedHough> hough = cv::GeneralizedHough::create(method);
if (method & GHT_ROTATION)
{
hough->set("maxAngle", 90.0);
hough->set("angleStep", 2.0);
}
cv::Mat positions;
hough->setTemplate(templ);
alg->setTemplate(templ);
TEST_CYCLE() hough->detect(edges, dx, dy, positions);
TEST_CYCLE() alg->detect(edges, dx, dy, positions);
CPU_SANITY_CHECK(positions);
}

262
modules/gpuimgproc/src/cuda/generalized_hough.cu
Unescape View File

@ -307,268 +307,6 @@ namespace cv { namespace gpu { namespace cudev
return totalCount;
}
////////////////////////////////////////////////////////////////////////
// Ballard_PosScale
__global__ void Ballard_PosScale_calcHist(const unsigned int* coordList, const float* thetaList,
PtrStep<short2> r_table, const int* r_sizes,
PtrStepi hist, const int rows, const int cols,
const float minScale, const float scaleStep, const int scaleRange,
const float idp, const float thetaScale)
{
const unsigned int coord = coordList[blockIdx.x];
float2 p;
p.x = (coord & 0xFFFF);
p.y = (coord >> 16) & 0xFFFF;
const float theta = thetaList[blockIdx.x];
const int n = __float2int_rn(theta * thetaScale);
const short2* r_row = r_table.ptr(n);
const int r_row_size = r_sizes[n];
for (int j = 0; j < r_row_size; ++j)
{
const float2 d = saturate_cast<float2>(r_row[j]);
for (int s = threadIdx.x; s < scaleRange; s += blockDim.x)
{
const float scale = minScale + s * scaleStep;
float2 c = p - scale * d;
c.x *= idp;
c.y *= idp;
if (c.x >= 0 && c.x < cols && c.y >= 0 && c.y < rows)
::atomicAdd(hist.ptr((s + 1) * (rows + 2) + __float2int_rn(c.y + 1)) + __float2int_rn(c.x + 1), 1);
}
}
}
void Ballard_PosScale_calcHist_gpu(const unsigned int* coordList, const float* thetaList, int pointsCount,
PtrStepSz<short2> r_table, const int* r_sizes,
PtrStepi hist, int rows, int cols,
float minScale, float scaleStep, int scaleRange,
float dp, int levels)
{
const dim3 block(256);
const dim3 grid(pointsCount);
const float idp = 1.0f / dp;
const float thetaScale = levels / (2.0f * CV_PI_F);
Ballard_PosScale_calcHist<<<grid, block>>>(coordList, thetaList,
r_table, r_sizes,
hist, rows, cols,
minScale, scaleStep, scaleRange,
idp, thetaScale);
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
}
__global__ void Ballard_PosScale_findPosInHist(const PtrStepi hist, const int rows, const int cols, const int scaleRange,
float4* out, int3* votes, const int maxSize,
const float minScale, const float scaleStep, const float dp, const int threshold)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x >= cols || y >= rows)
return;
for (int s = 0; s < scaleRange; ++s)
{
const float scale = minScale + s * scaleStep;
const int prevScaleIdx = (s) * (rows + 2);
const int curScaleIdx = (s + 1) * (rows + 2);
const int nextScaleIdx = (s + 2) * (rows + 2);
const int curVotes = hist(curScaleIdx + y + 1, x + 1);
if (curVotes > threshold &&
curVotes > hist(curScaleIdx + y + 1, x) &&
curVotes >= hist(curScaleIdx + y + 1, x + 2) &&
curVotes > hist(curScaleIdx + y, x + 1) &&
curVotes >= hist(curScaleIdx + y + 2, x + 1) &&
curVotes > hist(prevScaleIdx + y + 1, x + 1) &&
curVotes >= hist(nextScaleIdx + y + 1, x + 1))
{
const int ind = ::atomicAdd(&g_counter, 1);
if (ind < maxSize)
{
out[ind] = make_float4(x * dp, y * dp, scale, 0.0f);
votes[ind] = make_int3(curVotes, curVotes, 0);
}
}
}
}
int Ballard_PosScale_findPosInHist_gpu(PtrStepi hist, int rows, int cols, int scaleRange, float4* out, int3* votes, int maxSize,
float minScale, float scaleStep, float dp, int threshold)
{
void* counterPtr;
cudaSafeCall( cudaGetSymbolAddress(&counterPtr, g_counter) );
cudaSafeCall( cudaMemset(counterPtr, 0, sizeof(int)) );
const dim3 block(32, 8);
const dim3 grid(divUp(cols, block.x), divUp(rows, block.y));
cudaSafeCall( cudaFuncSetCacheConfig(Ballard_PosScale_findPosInHist, cudaFuncCachePreferL1) );
Ballard_PosScale_findPosInHist<<<grid, block>>>(hist, rows, cols, scaleRange, out, votes,
maxSize, minScale, scaleStep, dp, threshold);
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
int totalCount;
cudaSafeCall( cudaMemcpy(&totalCount, counterPtr, sizeof(int), cudaMemcpyDeviceToHost) );
totalCount = ::min(totalCount, maxSize);
return totalCount;
}
////////////////////////////////////////////////////////////////////////
// Ballard_PosRotation
__global__ void Ballard_PosRotation_calcHist(const unsigned int* coordList, const float* thetaList,
PtrStep<short2> r_table, const int* r_sizes,
PtrStepi hist, const int rows, const int cols,
const float minAngle, const float angleStep, const int angleRange,
const float idp, const float thetaScale)
{
const unsigned int coord = coordList[blockIdx.x];
float2 p;
p.x = (coord & 0xFFFF);
p.y = (coord >> 16) & 0xFFFF;
const float thetaVal = thetaList[blockIdx.x];
for (int a = threadIdx.x; a < angleRange; a += blockDim.x)
{
const float angle = (minAngle + a * angleStep) * (CV_PI_F / 180.0f);
float sinA, cosA;
sincosf(angle, &sinA, &cosA);
float theta = thetaVal - angle;
if (theta < 0)
theta += 2.0f * CV_PI_F;
const int n = __float2int_rn(theta * thetaScale);
const short2* r_row = r_table.ptr(n);
const int r_row_size = r_sizes[n];
for (int j = 0; j < r_row_size; ++j)
{
const float2 d = saturate_cast<float2>(r_row[j]);
const float2 dr = make_float2(d.x * cosA - d.y * sinA, d.x * sinA + d.y * cosA);
float2 c = make_float2(p.x - dr.x, p.y - dr.y);
c.x *= idp;
c.y *= idp;
if (c.x >= 0 && c.x < cols && c.y >= 0 && c.y < rows)
::atomicAdd(hist.ptr((a + 1) * (rows + 2) + __float2int_rn(c.y + 1)) + __float2int_rn(c.x + 1), 1);
}
}
}
void Ballard_PosRotation_calcHist_gpu(const unsigned int* coordList, const float* thetaList, int pointsCount,
PtrStepSz<short2> r_table, const int* r_sizes,
PtrStepi hist, int rows, int cols,
float minAngle, float angleStep, int angleRange,
float dp, int levels)
{
const dim3 block(256);
const dim3 grid(pointsCount);
const float idp = 1.0f / dp;
const float thetaScale = levels / (2.0f * CV_PI_F);
Ballard_PosRotation_calcHist<<<grid, block>>>(coordList, thetaList,
r_table, r_sizes,
hist, rows, cols,
minAngle, angleStep, angleRange,
idp, thetaScale);
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
}
__global__ void Ballard_PosRotation_findPosInHist(const PtrStepi hist, const int rows, const int cols, const int angleRange,
float4* out, int3* votes, const int maxSize,
const float minAngle, const float angleStep, const float dp, const int threshold)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x >= cols || y >= rows)
return;
for (int a = 0; a < angleRange; ++a)
{
const float angle = minAngle + a * angleStep;
const int prevAngleIdx = (a) * (rows + 2);
const int curAngleIdx = (a + 1) * (rows + 2);
const int nextAngleIdx = (a + 2) * (rows + 2);
const int curVotes = hist(curAngleIdx + y + 1, x + 1);
if (curVotes > threshold &&
curVotes > hist(curAngleIdx + y + 1, x) &&
curVotes >= hist(curAngleIdx + y + 1, x + 2) &&
curVotes > hist(curAngleIdx + y, x + 1) &&
curVotes >= hist(curAngleIdx + y + 2, x + 1) &&
curVotes > hist(prevAngleIdx + y + 1, x + 1) &&
curVotes >= hist(nextAngleIdx + y + 1, x + 1))
{
const int ind = ::atomicAdd(&g_counter, 1);
if (ind < maxSize)
{
out[ind] = make_float4(x * dp, y * dp, 1.0f, angle);
votes[ind] = make_int3(curVotes, 0, curVotes);
}
}
}
}
int Ballard_PosRotation_findPosInHist_gpu(PtrStepi hist, int rows, int cols, int angleRange, float4* out, int3* votes, int maxSize,
float minAngle, float angleStep, float dp, int threshold)
{
void* counterPtr;
cudaSafeCall( cudaGetSymbolAddress(&counterPtr, g_counter) );
cudaSafeCall( cudaMemset(counterPtr, 0, sizeof(int)) );
const dim3 block(32, 8);
const dim3 grid(divUp(cols, block.x), divUp(rows, block.y));
cudaSafeCall( cudaFuncSetCacheConfig(Ballard_PosRotation_findPosInHist, cudaFuncCachePreferL1) );
Ballard_PosRotation_findPosInHist<<<grid, block>>>(hist, rows, cols, angleRange, out, votes,
maxSize, minAngle, angleStep, dp, threshold);
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
int totalCount;
cudaSafeCall( cudaMemcpy(&totalCount, counterPtr, sizeof(int), cudaMemcpyDeviceToHost) );
totalCount = ::min(totalCount, maxSize);
return totalCount;
}
////////////////////////////////////////////////////////////////////////
// Guil_Full

1113
modules/gpuimgproc/src/generalized_hough.cpp
View File

File diff suppressed because it is too large Load Diff

12
modules/gpuimgproc/test/test_hough.cpp
Unescape View File

@ -193,7 +193,7 @@ PARAM_TEST_CASE(GeneralizedHough, cv::gpu::DeviceInfo, UseRoi)
{
};
GPU_TEST_P(GeneralizedHough, POSITION)
GPU_TEST_P(GeneralizedHough, Ballard)
{
const cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
cv::gpu::setDevice(devInfo.deviceID());
@ -218,16 +218,16 @@ GPU_TEST_P(GeneralizedHough, POSITION)
templ.copyTo(imageROI);
}
cv::Ptr<cv::gpu::GeneralizedHough> hough = cv::gpu::GeneralizedHough::create(cv::GeneralizedHough::GHT_POSITION);
hough->set("votesThreshold", 200);
cv::Ptr<cv::GeneralizedHoughBallard> alg = cv::gpu::createGeneralizedHoughBallard();
alg->setVotesThreshold(200);
hough->setTemplate(loadMat(templ, useRoi));
alg->setTemplate(loadMat(templ, useRoi));
cv::gpu::GpuMat d_pos;
hough->detect(loadMat(image, useRoi), d_pos);
alg->detect(loadMat(image, useRoi), d_pos);
std::vector<cv::Vec4f> pos;
hough->downloadResults(d_pos, pos);
d_pos.download(pos);
ASSERT_EQ(gold_count, pos.size());

119
modules/imgproc/include/opencv2/imgproc.hpp
Unescape View File

@ -688,39 +688,104 @@ public:
//! finds arbitrary template in the grayscale image using Generalized Hough Transform
class CV_EXPORTS GeneralizedHough : public Algorithm
{
public:
//! set template to search
virtual void setTemplate(InputArray templ, Point templCenter = Point(-1, -1)) = 0;
virtual void setTemplate(InputArray edges, InputArray dx, InputArray dy, Point templCenter = Point(-1, -1)) = 0;
//! find template on image
virtual void detect(InputArray image, OutputArray positions, OutputArray votes = noArray()) = 0;
virtual void detect(InputArray edges, InputArray dx, InputArray dy, OutputArray positions, OutputArray votes = noArray()) = 0;
//! Canny low threshold.
virtual void setCannyLowThresh(int cannyLowThresh) = 0;
virtual int getCannyLowThresh() const = 0;
//! Canny high threshold.
virtual void setCannyHighThresh(int cannyHighThresh) = 0;
virtual int getCannyHighThresh() const = 0;
//! Minimum distance between the centers of the detected objects.
virtual void setMinDist(double minDist) = 0;
virtual double getMinDist() const = 0;
//! Inverse ratio of the accumulator resolution to the image resolution.
virtual void setDp(double dp) = 0;
virtual double getDp() const = 0;
//! Maximal size of inner buffers.
virtual void setMaxBufferSize(int maxBufferSize) = 0;
virtual int getMaxBufferSize() const = 0;
};
//! Ballard, D.H. (1981). Generalizing the Hough transform to detect arbitrary shapes. Pattern Recognition 13 (2): 111-122.
//! Detects position only without traslation and rotation
class CV_EXPORTS GeneralizedHoughBallard : public GeneralizedHough
{
public:
//! R-Table levels.
virtual void setLevels(int levels) = 0;
virtual int getLevels() const = 0;
//! The accumulator threshold for the template centers at the detection stage. The smaller it is, the more false positions may be detected.
virtual void setVotesThreshold(int votesThreshold) = 0;
virtual int getVotesThreshold() const = 0;
};
//! Guil, N., González-Linares, J.M. and Zapata, E.L. (1999). Bidimensional shape detection using an invariant approach. Pattern Recognition 32 (6): 1025-1038.
class CV_EXPORTS GeneralizedHough : public Algorithm
//! Detects position, traslation and rotation
class CV_EXPORTS GeneralizedHoughGuil : public GeneralizedHough
{
public:
enum { GHT_POSITION = 0,
GHT_SCALE = 1,
GHT_ROTATION = 2
};
//! Angle difference in degrees between two points in feature.
virtual void setXi(double xi) = 0;
virtual double getXi() const = 0;
static Ptr<GeneralizedHough> create(int method);
//! Feature table levels.
virtual void setLevels(int levels) = 0;
virtual int getLevels() const = 0;
virtual ~GeneralizedHough();
//! Maximal difference between angles that treated as equal.
virtual void setAngleEpsilon(double angleEpsilon) = 0;
virtual double getAngleEpsilon() const = 0;
//! set template to search
void setTemplate(InputArray templ, int cannyThreshold = 100, Point templCenter = Point(-1, -1));
void setTemplate(InputArray edges, InputArray dx, InputArray dy, Point templCenter = Point(-1, -1));
//! Minimal rotation angle to detect in degrees.
virtual void setMinAngle(double minAngle) = 0;
virtual double getMinAngle() const = 0;
//! find template on image
void detect(InputArray image, OutputArray positions, OutputArray votes = cv::noArray(), int cannyThreshold = 100);
void detect(InputArray edges, InputArray dx, InputArray dy, OutputArray positions, OutputArray votes = cv::noArray());
//! Maximal rotation angle to detect in degrees.
virtual void setMaxAngle(double maxAngle) = 0;
virtual double getMaxAngle() const = 0;
void release();
//! Angle step in degrees.
virtual void setAngleStep(double angleStep) = 0;
virtual double getAngleStep() const = 0;
protected:
virtual void setTemplateImpl(const Mat& edges, const Mat& dx, const Mat& dy, Point templCenter) = 0;
virtual void detectImpl(const Mat& edges, const Mat& dx, const Mat& dy, OutputArray positions, OutputArray votes) = 0;
virtual void releaseImpl() = 0;
private:
Mat edges_;
Mat dx_;
Mat dy_;
//! Angle votes threshold.
virtual void setAngleThresh(int angleThresh) = 0;
virtual int getAngleThresh() const = 0;
//! Minimal scale to detect.
virtual void setMinScale(double minScale) = 0;
virtual double getMinScale() const = 0;
//! Maximal scale to detect.
virtual void setMaxScale(double maxScale) = 0;
virtual double getMaxScale() const = 0;
//! Scale step.
virtual void setScaleStep(double scaleStep) = 0;
virtual double getScaleStep() const = 0;
//! Scale votes threshold.
virtual void setScaleThresh(int scaleThresh) = 0;
virtual int getScaleThresh() const = 0;
//! Position votes threshold.
virtual void setPosThresh(int posThresh) = 0;
virtual int getPosThresh() const = 0;
};
@ -1355,6 +1420,14 @@ CV_EXPORTS_W double pointPolygonTest( InputArray contour, Point2f pt, bool measu
CV_EXPORTS Ptr<CLAHE> createCLAHE(double clipLimit = 40.0, Size tileGridSize = Size(8, 8));
//! Ballard, D.H. (1981). Generalizing the Hough transform to detect arbitrary shapes. Pattern Recognition 13 (2): 111-122.
//! Detects position only without traslation and rotation
CV_EXPORTS Ptr<GeneralizedHoughBallard> createGeneralizedHoughBallard();
//! Guil, N., González-Linares, J.M. and Zapata, E.L. (1999). Bidimensional shape detection using an invariant approach. Pattern Recognition 32 (6): 1025-1038.
//! Detects position, traslation and rotation
CV_EXPORTS Ptr<GeneralizedHoughGuil> createGeneralizedHoughGuil();
} // cv
#endif

1121
modules/imgproc/src/generalized_hough.cpp
View File

File diff suppressed because it is too large Load Diff

131
samples/gpu/generalized_hough.cpp
Unescape View File

@ -5,13 +5,12 @@
#include "opencv2/core.hpp"
#include "opencv2/core/utility.hpp"
#include "opencv2/imgproc.hpp"
#include "opencv2/gpu.hpp"
#include "opencv2/gpuimgproc.hpp"
#include "opencv2/highgui.hpp"
#include "opencv2/contrib.hpp"
using namespace std;
using namespace cv;
using cv::gpu::GpuMat;
static Mat loadImage(const string& name)
{
@ -29,8 +28,7 @@ int main(int argc, const char* argv[])
CommandLineParser cmd(argc, argv,
"{ image i | pic1.png | input image }"
"{ template t | templ.png | template image }"
"{ scale s | | estimate scale }"
"{ rotation r | | estimate rotation }"
"{ full | | estimate scale and rotation }"
"{ gpu | | use gpu version }"
"{ minDist | 100 | minimum distance between the centers of the detected objects }"
"{ levels | 360 | R-Table levels }"
@ -45,7 +43,7 @@ int main(int argc, const char* argv[])
"{ minAngle | 0 | minimal rotation angle to detect in degrees }"
"{ maxAngle | 360 | maximal rotation angle to detect in degrees }"
"{ angleStep | 1 | angle step in degrees }"
"{ maxSize | 1000 | maximal size of inner buffers }"
"{ maxBufSize | 1000 | maximal size of inner buffers }"
"{ help h ? | | print help message }"
);
@ -59,8 +57,7 @@ int main(int argc, const char* argv[])
const string templName = cmd.get<string>("template");
const string imageName = cmd.get<string>("image");
const bool estimateScale = cmd.has("scale");
const bool estimateRotation = cmd.has("rotation");
const bool full = cmd.has("full");
const bool useGpu = cmd.has("gpu");
const double minDist = cmd.get<double>("minDist");
const int levels = cmd.get<int>("levels");
@ -75,7 +72,7 @@ int main(int argc, const char* argv[])
const double minAngle = cmd.get<double>("minAngle");
const double maxAngle = cmd.get<double>("maxAngle");
const double angleStep = cmd.get<double>("angleStep");
const int maxSize = cmd.get<int>("maxSize");
const int maxBufSize = cmd.get<int>("maxBufSize");
if (!cmd.check())
{
@ -86,93 +83,69 @@ int main(int argc, const char* argv[])
Mat templ = loadImage(templName);
Mat image = loadImage(imageName);
int method = cv::GeneralizedHough::GHT_POSITION;
if (estimateScale)
method += cv::GeneralizedHough::GHT_SCALE;
if (estimateRotation)
method += cv::GeneralizedHough::GHT_ROTATION;
Ptr<GeneralizedHough> alg;
if (!full)
{
Ptr<GeneralizedHoughBallard> ballard = useGpu ? gpu::createGeneralizedHoughBallard() : createGeneralizedHoughBallard();
ballard->setMinDist(minDist);
ballard->setLevels(levels);
ballard->setDp(dp);
ballard->setMaxBufferSize(maxBufSize);
ballard->setVotesThreshold(votesThreshold);
alg = ballard;
}
else
{
Ptr<GeneralizedHoughGuil> guil = useGpu ? gpu::createGeneralizedHoughGuil() : createGeneralizedHoughGuil();
guil->setMinDist(minDist);
guil->setLevels(levels);
guil->setDp(dp);
guil->setMaxBufferSize(maxBufSize);
guil->setMinAngle(minAngle);
guil->setMaxAngle(maxAngle);
guil->setAngleStep(angleStep);
guil->setAngleThresh(angleThresh);
guil->setMinScale(minScale);
guil->setMaxScale(maxScale);
guil->setScaleStep(scaleStep);
guil->setScaleThresh(scaleThresh);
guil->setPosThresh(posThresh);
alg = guil;
}
vector<Vec4f> position;
cv::TickMeter tm;
TickMeter tm;
if (useGpu)
{
GpuMat d_templ(templ);
GpuMat d_image(image);
GpuMat d_position;
Ptr<gpu::GeneralizedHough> d_hough = gpu::GeneralizedHough::create(method);
d_hough->set("minDist", minDist);
d_hough->set("levels", levels);
d_hough->set("dp", dp);
d_hough->set("maxSize", maxSize);
if (estimateScale && estimateRotation)
{
d_hough->set("angleThresh", angleThresh);
d_hough->set("scaleThresh", scaleThresh);
d_hough->set("posThresh", posThresh);
}
else
{
d_hough->set("votesThreshold", votesThreshold);
}
if (estimateScale)
{
d_hough->set("minScale", minScale);
d_hough->set("maxScale", maxScale);
d_hough->set("scaleStep", scaleStep);
}
if (estimateRotation)
{
d_hough->set("minAngle", minAngle);
d_hough->set("maxAngle", maxAngle);
d_hough->set("angleStep", angleStep);
}
d_hough->setTemplate(d_templ);
gpu::GpuMat d_templ(templ);
gpu::GpuMat d_image(image);
gpu::GpuMat d_position;
alg->setTemplate(d_templ);
tm.start();
d_hough->detect(d_image, d_position);
d_hough->downloadResults(d_position, position);
alg->detect(d_image, d_position);
d_position.download(position);
tm.stop();
}
else
{
Ptr<GeneralizedHough> hough = GeneralizedHough::create(method);
hough->set("minDist", minDist);
hough->set("levels", levels);
hough->set("dp", dp);
if (estimateScale && estimateRotation)
{
hough->set("angleThresh", angleThresh);
hough->set("scaleThresh", scaleThresh);
hough->set("posThresh", posThresh);
hough->set("maxSize", maxSize);
}
else
{
hough->set("votesThreshold", votesThreshold);
}
if (estimateScale)
{
hough->set("minScale", minScale);
hough->set("maxScale", maxScale);
hough->set("scaleStep", scaleStep);
}
if (estimateRotation)
{
hough->set("minAngle", minAngle);
hough->set("maxAngle", maxAngle);
hough->set("angleStep", angleStep);
}
hough->setTemplate(templ);
alg->setTemplate(templ);
tm.start();
hough->detect(image, position);
alg->detect(image, position);
tm.stop();
}

Write Preview
Loading…
Cancel
Save