Update to 2.0.0 tree from current Fremantle build
[opencv] / src / cv / cvoptflowbm.cpp
diff --git a/src/cv/cvoptflowbm.cpp b/src/cv/cvoptflowbm.cpp
new file mode 100644 (file)
index 0000000..591cb6f
--- /dev/null
@@ -0,0 +1,303 @@
+/*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.
+//
+//
+//                        Intel License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000, Intel Corporation, 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 Intel Corporation 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 "_cv.h"
+
+
+static inline int cmpBlocks(const uchar* A, const uchar* B, int Bstep, CvSize blockSize )
+{
+    int x, s = 0;
+    for( ; blockSize.height--; A += blockSize.width, B += Bstep )
+    {
+        for( x = 0; x <= blockSize.width - 4; x += 4 )
+            s += std::abs(A[x] - B[x]) + std::abs(A[x+1] - B[x+1]) +
+                std::abs(A[x+2] - B[x+2]) + std::abs(A[x+3] - B[x+3]);
+        for( ; x < blockSize.width; x++ )
+            s += std::abs(A[x] - B[x]);
+    }
+    return s;
+}
+
+
+CV_IMPL void
+cvCalcOpticalFlowBM( const void* srcarrA, const void* srcarrB,
+                     CvSize blockSize, CvSize shiftSize,
+                     CvSize maxRange, int usePrevious,
+                     void* velarrx, void* velarry )
+{
+    CvMat stubA, *srcA = cvGetMat( srcarrA, &stubA );
+    CvMat stubB, *srcB = cvGetMat( srcarrB, &stubB );
+
+    CvMat stubx, *velx = cvGetMat( velarrx, &stubx );
+    CvMat stuby, *vely = cvGetMat( velarry, &stuby );
+
+    if( !CV_ARE_TYPES_EQ( srcA, srcB ))
+        CV_Error( CV_StsUnmatchedFormats, "Source images have different formats" );
+
+    if( !CV_ARE_TYPES_EQ( velx, vely ))
+        CV_Error( CV_StsUnmatchedFormats, "Destination images have different formats" );
+
+    CvSize velSize =
+    {
+        (srcA->width - blockSize.width)/shiftSize.width,
+        (srcA->height - blockSize.height)/shiftSize.height
+    };
+
+    if( !CV_ARE_SIZES_EQ( srcA, srcB ) ||
+        !CV_ARE_SIZES_EQ( velx, vely ) ||
+        velx->width != velSize.width ||
+        vely->height != velSize.height )
+        CV_Error( CV_StsUnmatchedSizes, "" );
+
+    if( CV_MAT_TYPE( srcA->type ) != CV_8UC1 ||
+        CV_MAT_TYPE( velx->type ) != CV_32FC1 )
+        CV_Error( CV_StsUnsupportedFormat, "Source images must have 8uC1 type and "
+                                           "destination images must have 32fC1 type" );
+
+    if( srcA->step != srcB->step || velx->step != vely->step )
+        CV_Error( CV_BadStep, "two source or two destination images have different steps" );
+
+    const int SMALL_DIFF=2;
+    const int BIG_DIFF=128;
+    
+    // scanning scheme coordinates
+    cv::vector<CvPoint> _ss((2 * maxRange.width + 1) * (2 * maxRange.height + 1));
+    CvPoint* ss = &_ss[0];
+    int ss_count = 0;
+
+    int blWidth = blockSize.width, blHeight = blockSize.height;
+    int blSize = blWidth*blHeight;
+    int acceptLevel = blSize * SMALL_DIFF;
+    int escapeLevel = blSize * BIG_DIFF;
+    
+    int i, j;
+
+    cv::vector<uchar> _blockA(cvAlign(blSize + 16, 16));
+    uchar* blockA = (uchar*)cvAlignPtr(&_blockA[0], 16);
+    
+    // Calculate scanning scheme
+    int min_count = 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( i = 0; i < min_count; i++ )
+    {
+        // four cycles along sides
+        int x = -i-1, y = x;
+        
+        // upper side
+        for( j = -i; j <= i + 1; j++, ss_count++ )
+        {
+            ss[ss_count].x = ++x;
+            ss[ss_count].y = y;
+        }
+        
+        // right side
+        for( j = -i; j <= i + 1; j++, ss_count++ )
+        {
+            ss[ss_count].x = x;
+            ss[ss_count].y = ++y;
+        }
+        
+        // bottom side
+        for( j = -i; j <= i + 1; j++, ss_count++ )
+        {
+            ss[ss_count].x = --x;
+            ss[ss_count].y = y;
+        }
+        
+        // left side
+        for( j = -i; j <= i + 1; j++, ss_count++ )
+        {
+            ss[ss_count].x = x;
+            ss[ss_count].y = --y;
+        }
+    }
+    
+    // the rest part
+    if( maxRange.width < maxRange.height )
+    {
+        int xleft = -min_count;
+        
+        // cycle by neighbor rings
+        for( i = min_count; i < maxRange.height; i++ )
+        {
+            // two cycles by x
+            int y = -(i + 1);
+            int x = xleft;
+            
+            // upper side
+            for( j = -maxRange.width; j <= maxRange.width; j++, ss_count++, x++ )
+            {
+                ss[ss_count].x = x;
+                ss[ss_count].y = y;
+            }
+            
+            x = xleft;
+            y = -y;
+            // bottom side
+            for( j = -maxRange.width; j <= maxRange.width; j++, ss_count++, x++ )
+            {
+                ss[ss_count].x = x;
+                ss[ss_count].y = y;
+            }
+        }
+    }
+    else if( maxRange.width > maxRange.height )
+    {
+        int yupper = -min_count;
+        
+        // cycle by neighbor rings
+        for( i = min_count; i < maxRange.width; i++ )
+        {
+            // two cycles by y
+            int x = -(i + 1);
+            int y = yupper;
+            
+            // left side
+            for( j = -maxRange.height; j <= maxRange.height; j++, ss_count++, y++ )
+            {
+                ss[ss_count].x = x;
+                ss[ss_count].y = y;
+            }
+            
+            y = yupper;
+            x = -x;
+            // right side
+            for( j = -maxRange.height; j <= maxRange.height; j++, ss_count++, y++ )
+            {
+                ss[ss_count].x = x;
+                ss[ss_count].y = y;
+            }
+        }
+    }
+
+    int maxX = srcB->cols - blockSize.width, maxY = srcB->rows - blockSize.height;
+    const uchar* Adata = srcA->data.ptr;
+    const uchar* Bdata = srcB->data.ptr;
+    int Astep = srcA->step, Bstep = srcB->step;
+    
+    // compute the flow
+    for( i = 0; i < velx->rows; i++ )
+    {
+        float* vx = (float*)(velx->data.ptr + velx->step*i);
+        float* vy = (float*)(vely->data.ptr + vely->step*i);
+        
+        for( j = 0; j < velx->cols; j++ )
+        {
+            int X1 = j*shiftSize.width, Y1 = i*shiftSize.height, X2, Y2;
+            int offX = 0, offY = 0;
+            
+            if( usePrevious )
+            {
+                offX = cvRound(vx[j]);
+                offY = cvRound(vy[j]);
+            }
+
+            int k;
+            for( k = 0; k < blHeight; k++ )
+                memcpy( blockA + k*blWidth, Adata + Astep*(Y1 + k) + X1, blWidth );
+            
+            X2 = X1 + offX;
+            Y2 = Y1 + offY;
+            int dist = INT_MAX;
+            if( 0 <= X2 && X2 <= maxX && 0 <= Y2 && Y2 <= maxY )
+                dist = cmpBlocks( blockA, Bdata + Bstep*Y2 + X2, Bstep, blockSize );
+
+            int countMin = 1;
+            int sumx = offX, sumy = offY;
+
+            if( dist > acceptLevel )
+            {
+                // do brute-force search
+                for( k = 0; k < ss_count; k++ )
+                {
+                    int dx = offX + ss[k].x;
+                    int dy = offY + ss[k].y;
+                    X2 = X1 + dx;
+                    Y2 = Y1 + dy;
+                    
+                    if( !(0 <= X2 && X2 <= maxX && 0 <= Y2 && Y2 <= maxY) )
+                        continue;
+                    
+                    int tmpDist = cmpBlocks( blockA, Bdata + Bstep*Y2 + X2, Bstep, 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;
+                }
+            }
+            
+            vx[j] = (float)sumx/countMin;
+            vy[j] = (float)sumy/countMin;
+        } 
+    }
+}
+
+/* End of file. */