Added HoughLinesP OCL implementation

modules/imgproc/src/hough.cpp

@@ -652,6 +652,9 @@ HoughLinesProbabilistic( Mat& image,
         }
     }
 }
+
+#ifdef HAVE_OPENCL
+
 static bool ocl_makePointsList(InputArray _src, OutputArray _pointsList, InputOutputArray _counters)
 {
     UMat src = _src.getUMat();
@@ -660,16 +663,16 @@ static bool ocl_makePointsList(InputArray _src, OutputArray _pointsList, InputOu
     UMat counters = _counters.getUMat();
     ocl::Device dev = ocl::Device::getDefault();
 
-    const int pixelsPerWI = 16;
-    int workgroup_size = min((int) dev.maxWorkGroupSize(), (src.cols + pixelsPerWI - 1)/pixelsPerWI);
-    ocl::Kernel pointListKernel("make_point_list", ocl::imgproc::hough_lines_oclsrc, 
+    const int pixPerWI = 16;
+    int workgroup_size = min((int) dev.maxWorkGroupSize(), (src.cols + pixPerWI - 1)/pixPerWI);
+    ocl::Kernel pointListKernel("make_point_list", ocl::imgproc::hough_lines_oclsrc,
                                 format("-D MAKE_POINTS_LIST -D GROUP_SIZE=%d -D LOCAL_SIZE=%d", workgroup_size, src.cols));
     if (pointListKernel.empty())
         return false;
 
     pointListKernel.args(ocl::KernelArg::ReadOnly(src), ocl::KernelArg::WriteOnlyNoSize(pointsList),
                          ocl::KernelArg::PtrWriteOnly(counters));
-    
+
     size_t localThreads[2]  = { workgroup_size, 1 };
     size_t globalThreads[2] = { workgroup_size, src.rows };
 
@@ -685,12 +688,12 @@ static bool ocl_fillAccum(InputArray _pointsList, OutputArray _accum, int total_
 
     float irho = (float) (1 / rho);
     int workgroup_size = min((int) dev.maxWorkGroupSize(), total_points);
-    
+
     ocl::Kernel fillAccumKernel;
     size_t localThreads[2];
     size_t globalThreads[2];
 
-    int local_memory_needed = (numrho + 2)*sizeof(int);
+    size_t local_memory_needed = (numrho + 2)*sizeof(int);
     if (local_memory_needed > dev.localMemSize())
     {
         accum.setTo(Scalar::all(0));
@@ -717,7 +720,7 @@ static bool ocl_fillAccum(InputArray _pointsList, OutputArray _accum, int total_
     }
 }
 
-static bool ocl_HoughLines(InputArray _src, OutputArray _lines, double rho, double theta, int threshold,  
+static bool ocl_HoughLines(InputArray _src, OutputArray _lines, double rho, double theta, int threshold,
                            double min_theta, double max_theta)
 {
     CV_Assert(_src.type() == CV_8UC1);
@@ -732,7 +735,7 @@ static bool ocl_HoughLines(InputArray _src, OutputArray _lines, double rho, doub
     UMat src = _src.getUMat();
     int numangle = cvRound((max_theta - min_theta) / theta);
     int numrho = cvRound(((src.cols + src.rows) * 2 + 1) / rho);
-    
+
     UMat pointsList;
     UMat counters(1, 2, CV_32SC1, Scalar::all(0));
 
@@ -766,7 +769,7 @@ static bool ocl_HoughLines(InputArray _src, OutputArray _lines, double rho, doub
     size_t globalThreads[2] = { (numrho + pixPerWI - 1)/pixPerWI, numangle };
     if (!getLinesKernel.run(2, globalThreads, NULL, false))
         return false;
-    
+
     int total_lines = min(counters.getMat(ACCESS_READ).at<int>(0, 1), linesMax);
     if (total_lines > 0)
         _lines.assign(lines.rowRange(Range(0, total_lines)));
@@ -775,23 +778,67 @@ static bool ocl_HoughLines(InputArray _src, OutputArray _lines, double rho, doub
     return true;
 }
 
-static bool ocl_HoughLinesP(InputArray _src, OutputArray _lines, double rho, double theta, int threshold,  
+static bool ocl_HoughLinesP(InputArray _src, OutputArray _lines, double rho, double theta, int threshold,
                            double minLineLength, double maxGap)
 {
     CV_Assert(_src.type() == CV_8UC1);
-    
+
     UMat src = _src.getUMat();
+    int numangle = cvRound(CV_PI / theta);
+    int numrho = cvRound(((src.cols + src.rows) * 2 + 1) / rho);
+
+    UMat pointsList;
+    UMat counters(1, 2, CV_32SC1, Scalar::all(0));
+
+    if (!ocl_makePointsList(src, pointsList, counters))
+        return false;
 
-    return false;
+    int total_points = counters.getMat(ACCESS_READ).at<int>(0, 0);
+    if (total_points <= 0)
+    {
+        _lines.assign(UMat(0,0,CV_32SC4));
+        return true;
+    }
+
+    UMat accum;
+    if (!ocl_fillAccum(pointsList, accum, total_points, rho, theta, numrho, numangle))
+        return false;
+
+    ocl::Kernel getLinesKernel("get_lines", ocl::imgproc::hough_lines_oclsrc,
+                               format("-D GET_LINES_PROBABOLISTIC"));
+    if (getLinesKernel.empty())
+        return false;
+
+    // TODO: investigate other strategies to choose linesMax
+    int linesMax = min(total_points*numangle/threshold, 4096);
+    UMat lines(linesMax, 1, CV_32SC4);
+
+    getLinesKernel.args(ocl::KernelArg::ReadOnly(accum), ocl::KernelArg::ReadOnly(src),
+                        ocl::KernelArg::WriteOnlyNoSize(lines), ocl::KernelArg::PtrWriteOnly(counters),
+                        linesMax, threshold, (int) minLineLength, (int) maxGap, (float) rho, (float) theta);
+
+    size_t globalThreads[2] = { numrho, numangle };
+    if (!getLinesKernel.run(2, globalThreads, NULL, false))
+        return false;
+
+    int total_lines = min(counters.getMat(ACCESS_READ).at<int>(0, 1), linesMax);
+    if (total_lines > 0)
+        _lines.assign(lines.rowRange(Range(0, total_lines)));
+    else
+        _lines.assign(UMat(0,0,CV_32SC4));
+
+    return true;
 }
 
+#endif /* HAVE_OPENCL */
+
 }
 
 void cv::HoughLines( InputArray _image, OutputArray _lines,
                     double rho, double theta, int threshold,
                     double srn, double stn, double min_theta, double max_theta )
 {
-    CV_OCL_RUN(srn == 0 && stn == 0 && _image.isUMat() && _lines.isUMat(), 
+    CV_OCL_RUN(srn == 0 && stn == 0 && _image.isUMat() && _lines.isUMat(),
                ocl_HoughLines(_image, _lines, rho, theta, threshold, min_theta, max_theta));
 
     Mat image = _image.getMat();
@@ -810,7 +857,8 @@ void cv::HoughLinesP(InputArray _image, OutputArray _lines,
                      double rho, double theta, int threshold,
                      double minLineLength, double maxGap )
 {
-    CV_OCL_RUN(_image.isUMat() && _lines.isUMat(), ocl_HoughLinesP(_image, _lines, rho, theta, threshold, minLineLength, maxGap));
+    CV_OCL_RUN(_image.isUMat() && _lines.isUMat(),
+               ocl_HoughLinesP(_image, _lines, rho, theta, threshold, minLineLength, maxGap));
 
     Mat image = _image.getMat();
     std::vector<Vec4i> lines;

modules/imgproc/src/opencl/hough_lines.cl

@@ -12,7 +12,7 @@ __kernel void make_point_list(__global const uchar * src_ptr, int src_step, int
 {
     int x = get_local_id(0);
     int y = get_group_id(1);
-    
+
     __local int l_index, l_offset;
     __local int l_points[LOCAL_SIZE];
     __global const uchar * src = src_ptr + mad24(y, src_step, src_offset);
@@ -37,12 +37,12 @@ __kernel void make_point_list(__global const uchar * src_ptr, int src_step, int
     }
 
     barrier(CLK_LOCAL_MEM_FENCE);
-    
+
     if (x == 0)
         l_offset = atomic_add(global_offset, l_index);
 
     barrier(CLK_LOCAL_MEM_FENCE);
-    
+
     list += l_offset;
     for (int i=x; i < l_index; i+=GROUP_SIZE)
     {
@@ -53,8 +53,8 @@ __kernel void make_point_list(__global const uchar * src_ptr, int src_step, int
 #elif defined FILL_ACCUM_GLOBAL
 
 __kernel void fill_accum_global(__global const uchar * list_ptr, int list_step, int list_offset,
-                         __global uchar * accum_ptr, int accum_step, int accum_offset, int accum_rows, int accum_cols,
-                         int total_points, float irho, float theta, int numrho, int numangle)
+                                __global uchar * accum_ptr, int accum_step, int accum_offset, int accum_rows, int accum_cols,
+                                int total_points, float irho, float theta, int numrho, int numangle)
 {
     int theta_idx = get_global_id(1);
     int count_idx = get_global_id(0);
@@ -90,7 +90,7 @@ __kernel void fill_accum_local(__global const uchar * list_ptr, int list_step, i
 {
     int theta_idx = get_group_id(1);
     int count_idx = get_local_id(0);
-    
+
     if (theta_idx > 0 && theta_idx < numangle + 1)
     {
         float cosVal;
@@ -136,7 +136,7 @@ __kernel void fill_accum_local(__global const uchar * list_ptr, int list_step, i
 #define ACCUM(ptr) *((__global int*)(ptr))
 
 __kernel void get_lines(__global uchar * accum_ptr, int accum_step, int accum_offset, int accum_rows, int accum_cols,
-                         __global uchar * lines_ptr, int lines_step, int lines_offset, __global int* lines_index_ptr, 
+                         __global uchar * lines_ptr, int lines_step, int lines_offset, __global int* lines_index_ptr,
                          int linesMax, int threshold, float rho, float theta)
 {
     int x0 = get_global_id(0);
@@ -148,7 +148,7 @@ __kernel void get_lines(__global uchar * accum_ptr, int accum_step, int accum_of
         __global uchar* accum = accum_ptr + mad24(y+1, accum_step, mad24(x0+1, (int) sizeof(int), accum_offset));
         __global float2* lines = (__global float2*)(lines_ptr + lines_offset);
         __global int* lines_index = lines_index_ptr + 1;
-    
+
         for (int x=x0; x<accum_cols-2; x+=gl_size)
         {
             int curVote = ACCUM(accum);
@@ -172,4 +172,169 @@ __kernel void get_lines(__global uchar * accum_ptr, int accum_step, int accum_of
     }
 }
 
+#elif GET_LINES_PROBABOLISTIC
+
+#define ACCUM(ptr) *((__global int*)(ptr))
+
+__kernel void get_lines(__global const uchar * accum_ptr, int accum_step, int accum_offset, int accum_rows, int accum_cols,
+                        __global const uchar * src_ptr, int src_step, int src_offset, int src_rows, int src_cols,
+                        __global uchar * lines_ptr, int lines_step, int lines_offset, __global int* lines_index_ptr,
+                        int linesMax, int threshold, int lineLength, int lineGap, float rho, float theta)
+{
+    int x = get_global_id(0);
+    int y = get_global_id(1);
+
+    __global uchar* accum = accum_ptr + mad24(y+1, accum_step, mad24(x+1, (int) sizeof(int), accum_offset));
+    __global int4* lines = (__global int4*)(lines_ptr + lines_offset);
+    __global int* lines_index = lines_index_ptr + 1;
+
+    int curVote = ACCUM(accum);
+
+    if (curVote >= threshold &&
+        curVote > ACCUM(accum - accum_step - sizeof(int)) &&
+        curVote > ACCUM(accum - accum_step) &&
+        curVote > ACCUM(accum - accum_step + sizeof(int)) &&
+        curVote > ACCUM(accum - sizeof(int)) &&
+        curVote > ACCUM(accum + sizeof(int)) &&
+        curVote > ACCUM(accum + accum_step - sizeof(int)) &&
+        curVote > ACCUM(accum + accum_step) &&
+        curVote > ACCUM(accum + accum_step + sizeof(int)))
+    {
+        const float radius = (x - (accum_cols - 2 - 1) * 0.5f) * rho;
+        const float angle = y * theta;
+
+        float cosa;
+        float sina = sincos(angle, &cosa);
+
+        float2 p0 = (float2)(cosa * radius, sina * radius);
+        float2 dir = (float2)(-sina, cosa);
+
+        float2 pb[4] = { (float2)(-1, -1), (float2)(-1, -1), (float2)(-1, -1), (float2)(-1, -1) };
+        float a;
+
+        if (dir.x != 0)
+        {
+            a = -p0.x / dir.x;
+            pb[0].x = 0;
+            pb[0].y = p0.y + a * dir.y;
+
+            a = (src_cols - 1 - p0.x) / dir.x;
+            pb[1].x = src_cols - 1;
+            pb[1].y = p0.y + a * dir.y;
+        }
+        if (dir.y != 0)
+        {
+            a = -p0.y / dir.y;
+            pb[2].x = p0.x + a * dir.x;
+            pb[2].y = 0;
+
+            a = (src_rows - 1 - p0.y) / dir.y;
+            pb[3].x = p0.x + a * dir.x;
+            pb[3].y = src_rows - 1;
+        }
+
+        if (pb[0].x == 0 && (pb[0].y >= 0 && pb[0].y < src_rows))
+        {
+            p0 = pb[0];
+            if (dir.x < 0)
+                dir = -dir;
+        }
+        else if (pb[1].x == src_cols - 1 && (pb[0].y >= 0 && pb[0].y < src_rows))
+        {
+            p0 = pb[1];
+            if (dir.x > 0)
+                dir = -dir;
+        }
+        else if (pb[2].y == 0 && (pb[2].x >= 0 && pb[2].x < src_cols))
+        {
+            p0 = pb[2];
+            if (dir.y < 0)
+                dir = -dir;
+        }
+        else if (pb[3].y == src_rows - 1 && (pb[3].x >= 0 && pb[3].x < src_cols))
+        {
+            p0 = pb[3];
+            if (dir.y > 0)
+                dir = -dir;
+        }
+
+        float2 d;
+        if (fabs(dir.x) > fabs(dir.y))
+        {
+            d.x = dir.x > 0 ? 1 : -1;
+            d.y = dir.y / fabs(dir.x);
+        }
+        else
+        {
+            d.x = dir.x / fabs(dir.y);
+            d.y = dir.y > 0 ? 1 : -1;
+        }
+
+        float2 line_end[2];
+        int gap;
+        bool inLine = false;
+
+        float2 p1 = p0;
+        if (p1.x < 0 || p1.x >= src_cols || p1.y < 0 || p1.y >= src_rows)
+            return;
+
+        for (;;)
+        {
+            if (*(src_ptr + mad24(p1.y, src_step, p1.x + src_offset)))
+            {
+                gap = 0;
+
+                if (!inLine)
+                {
+                    line_end[0] = p1;
+                    line_end[1] = p1;
+                    inLine = true;
+                }
+                else
+                {
+                    line_end[1] = p1;
+                }
+            }
+            else if (inLine)
+            {
+                if (++gap > lineGap)
+                {
+                    bool good_line = fabs(line_end[1].x - line_end[0].x) >= lineLength ||
+                                     fabs(line_end[1].y - line_end[0].y) >= lineLength;
+
+                    if (good_line)
+                    {
+                        int index = atomic_inc(lines_index);
+                        if (index < linesMax)
+                            lines[index] = (int4)(line_end[0].x, line_end[0].y, line_end[1].x, line_end[1].y);
+                    }
+
+                    gap = 0;
+                    inLine = false;
+                }
+            }
+
+            p1 = p1 + d;
+            if (p1.x < 0 || p1.x >= src_cols || p1.y < 0 || p1.y >= src_rows)
+            {
+                if (inLine)
+                {
+                    bool good_line = fabs(line_end[1].x - line_end[0].x) >= lineLength ||
+                                     fabs(line_end[1].y - line_end[0].y) >= lineLength;
+
+                    if (good_line)
+                    {
+                        int index = atomic_inc(lines_index);
+                        if (index < linesMax)
+                            lines[index] = (int4)(line_end[0].x, line_end[0].y, line_end[1].x, line_end[1].y);
+                    }
+
+                }
+                break;
+            }
+        }
+
+    }
+}
+
 #endif

modules/imgproc/test/ocl/test_houghlines.cpp

@@ -15,17 +15,18 @@ namespace ocl {
 
 struct Vec2fComparator
 {
-    bool operator()(const cv::Vec2f& a, const cv::Vec2f b) const
+    bool operator()(const Vec2f& a, const Vec2f b) const
     {
         if(a[0] != b[0]) return a[0] < b[0];
         else return a[1] < b[1];
     }
 };
 
-PARAM_TEST_CASE(HoughLinesTestBase, double, double, int)
+/////////////////////////////// HoughLines ////////////////////////////////////
+
+PARAM_TEST_CASE(HoughLines, double, double, int)
 {
-    double rhoStep;
-    double thetaStep;
+    double rhoStep, thetaStep;
     int threshold;
 
     Size src_size;
@@ -50,7 +51,7 @@ PARAM_TEST_CASE(HoughLinesTestBase, double, double, int)
         line(src, Point(100, 0), Point(100, 200), Scalar::all(255), 1);
         line(src, Point(200, 0), Point(200, 200), Scalar::all(255), 1);
         line(src, Point(400, 0), Point(400, 200), Scalar::all(255), 1);
-        
+
         src.copyTo(usrc);
     }
 
@@ -65,7 +66,7 @@ PARAM_TEST_CASE(HoughLinesTestBase, double, double, int)
     virtual void Near(double eps = 0.)
     {
         EXPECT_EQ(dst.size(), udst.size());
-        
+
         if (dst.total() > 0)
         {
             Mat lines_cpu, lines_gpu;
@@ -80,8 +81,6 @@ PARAM_TEST_CASE(HoughLinesTestBase, double, double, int)
     }
 };
 
-typedef HoughLinesTestBase HoughLines;
-
 OCL_TEST_P(HoughLines, RealImage)
 {
     readRealTestData();
@@ -105,10 +104,81 @@ OCL_TEST_P(HoughLines, GeneratedImage)
     }
 }
 
+/////////////////////////////// HoughLinesP ///////////////////////////////////
+
+PARAM_TEST_CASE(HoughLinesP, int, double, double)
+{
+    double rhoStep, thetaStep, minLineLength, maxGap;
+    int threshold;
+
+    Size src_size;
+    Mat src, dst;
+    UMat usrc, udst;
+
+    virtual void SetUp()
+    {
+        rhoStep = 1.0;
+        thetaStep = CV_PI / 180;
+        threshold = GET_PARAM(0);
+        minLineLength = GET_PARAM(1);
+        maxGap = GET_PARAM(2);
+    }
+
+    virtual void readRealTestData()
+    {
+        Mat img = readImage("shared/pic5.png", IMREAD_GRAYSCALE);
+        Canny(img, src, 50, 200, 3);
+
+        src.copyTo(usrc);
+    }
+
+    virtual void Near(double eps = 0.)
+    {
+        Mat lines_gpu = udst.getMat(ACCESS_READ);
+
+        if (dst.total() > 0 && lines_gpu.total() > 0)
+        {
+            Mat result_cpu(src.size(), CV_8UC1, Scalar::all(0));
+            Mat result_gpu(src.size(), CV_8UC1, Scalar::all(0));
+
+            MatConstIterator_<Vec4i> it = dst.begin<Vec4i>(), end = dst.end<Vec4i>();
+            for ( ; it != end; it++)
+            {
+                Vec4i p = *it;
+                line(result_cpu, Point(p[0], p[1]), Point(p[2], p[3]), Scalar(255));
+            }
+
+            it = lines_gpu.begin<Vec4i>(), end = lines_gpu.end<Vec4i>();
+            for ( ; it != end; it++)
+            {
+                Vec4i p = *it;
+                line(result_gpu, Point(p[0], p[1]), Point(p[2], p[3]), Scalar(255));
+            }
+
+            EXPECT_MAT_SIMILAR(result_cpu, result_gpu, eps);
+        }
+    }
+};
+
+
+OCL_TEST_P(HoughLinesP, RealImage)
+{
+    readRealTestData();
+
+    OCL_OFF(cv::HoughLinesP(src, dst, rhoStep, thetaStep, threshold, minLineLength, maxGap));
+    OCL_ON(cv::HoughLinesP(usrc, udst, rhoStep, thetaStep, threshold, minLineLength, maxGap));
+
+    Near(0.2);
+}
+
 OCL_INSTANTIATE_TEST_CASE_P(Imgproc, HoughLines, Combine(Values(1, 0.5),                        // rhoStep
                                                          Values(CV_PI / 180.0, CV_PI / 360.0),  // thetaStep
                                                          Values(80, 150)));                     // threshold
 
+OCL_INSTANTIATE_TEST_CASE_P(Imgproc, HoughLinesP, Combine(Values(100, 150),                     // threshold
+                                                          Values(50, 100),                      // minLineLength
+                                                          Values(5, 10)));                      // maxLineGap
+
 } } // namespace cvtest::ocl
 
 #endif // HAVE_OPENCL