Open Source Computer Vision Library https://opencv.org/
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

356 lines
12 KiB

/*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 "test_precomp.hpp"
#include "opencv2/video/tracking.hpp"
#include <string>
#include <iostream>
#include <fstream>
#include <iterator>
#include <limits>
using namespace cv;
using namespace std;
class CV_OptFlowTest : public cvtest::BaseTest
{
public:
CV_OptFlowTest();
~CV_OptFlowTest();
protected:
void run(int);
bool runDense(const Point& shift = Point(3, 0));
bool runSparse();
};
CV_OptFlowTest::CV_OptFlowTest() {}
CV_OptFlowTest::~CV_OptFlowTest() {}
Mat copnvert2flow(const Mat& velx, const Mat& vely)
{
Mat flow(velx.size(), CV_32FC2);
for(int y = 0 ; y < flow.rows; ++y)
for(int x = 0 ; x < flow.cols; ++x)
flow.at<Point2f>(y, x) = Point2f(velx.at<float>(y, x), vely.at<float>(y, x));
return flow;
}
void calcOpticalFlowLK( const Mat& prev, const Mat& curr, Size winSize, Mat& flow )
{
Mat velx(prev.size(), CV_32F), vely(prev.size(), CV_32F);
CvMat cvvelx = velx; CvMat cvvely = vely;
CvMat cvprev = prev; CvMat cvcurr = curr;
cvCalcOpticalFlowLK( &cvprev, &cvcurr, winSize, &cvvelx, &cvvely );
flow = copnvert2flow(velx, vely);
}
void calcOpticalFlowBM( const Mat& prev, const Mat& curr, Size bSize, Size shiftSize, Size maxRange, int usePrevious, Mat& flow )
{
Size sz((curr.cols - bSize.width + shiftSize.width)/shiftSize.width, (curr.rows - bSize.height + shiftSize.height)/shiftSize.height);
Mat velx(sz, CV_32F), vely(sz, CV_32F);
CvMat cvvelx = velx; CvMat cvvely = vely;
CvMat cvprev = prev; CvMat cvcurr = curr;
cvCalcOpticalFlowBM( &cvprev, &cvcurr, bSize, shiftSize, maxRange, usePrevious, &cvvelx, &cvvely);
flow = copnvert2flow(velx, vely);
}
void calcOpticalFlowHS( const Mat& prev, const Mat& curr, int usePrevious, double lambda, TermCriteria criteria, Mat& flow)
{
Mat velx(prev.size(), CV_32F), vely(prev.size(), CV_32F);
CvMat cvvelx = velx; CvMat cvvely = vely;
CvMat cvprev = prev; CvMat cvcurr = curr;
cvCalcOpticalFlowHS( &cvprev, &cvcurr, usePrevious, &cvvelx, &cvvely, lambda, criteria );
flow = copnvert2flow(velx, vely);
}
void calcAffineFlowPyrLK( const Mat& prev, const Mat& curr,
const vector<Point2f>& prev_features, vector<Point2f>& curr_features,
vector<uchar>& status, vector<float>& track_error, vector<float>& matrices,
TermCriteria criteria = TermCriteria(TermCriteria::COUNT+TermCriteria::EPS,30, 0.01),
Size win_size = Size(15, 15), int level = 3, int flags = 0)
{
CvMat cvprev = prev;
CvMat cvcurr = curr;
size_t count = prev_features.size();
curr_features.resize(count);
status.resize(count);
track_error.resize(count);
matrices.resize(count * 6);
cvCalcAffineFlowPyrLK( &cvprev, &cvcurr, 0, 0,
(const CvPoint2D32f*)&prev_features[0], (CvPoint2D32f*)&curr_features[0], &matrices[0],
(int)count, win_size, level, (char*)&status[0], &track_error[0], criteria, flags );
}
double showFlowAndCalcError(const string& name, const Mat& gray, const Mat& flow,
const Rect& where, const Point& d,
bool showImages = false, bool writeError = false)
{
const int mult = 16;
if (showImages)
{
Mat tmp, cflow;
resize(gray, tmp, gray.size() * mult, 0, 0, INTER_NEAREST);
cvtColor(tmp, cflow, CV_GRAY2BGR);
const float m2 = 0.3f;
const float minVel = 0.1f;
for(int y = 0; y < flow.rows; ++y)
for(int x = 0; x < flow.cols; ++x)
{
Point2f f = flow.at<Point2f>(y, x);
if (f.x * f.x + f.y * f.y > minVel * minVel)
{
Point p1 = Point(x, y) * mult;
Point p2 = Point(cvRound((x + f.x*m2) * mult), cvRound((y + f.y*m2) * mult));
line(cflow, p1, p2, CV_RGB(0, 255, 0));
circle(cflow, Point(x, y) * mult, 2, CV_RGB(255, 0, 0));
}
}
rectangle(cflow, (where.tl() + d) * mult, (where.br() + d - Point(1,1)) * mult, CV_RGB(0, 0, 255));
namedWindow(name, 1); imshow(name, cflow);
}
double angle = atan2((float)d.y, (float)d.x);
double error = 0;
bool all = true;
Mat inner = flow(where);
for(int y = 0; y < inner.rows; ++y)
for(int x = 0; x < inner.cols; ++x)
{
const Point2f f = inner.at<Point2f>(y, x);
if (f.x == 0 && f.y == 0)
continue;
all = false;
double a = atan2(f.y, f.x);
error += fabs(angle - a);
}
double res = all ? numeric_limits<double>::max() : error / (inner.cols * inner.rows);
if (writeError)
cout << "Error " + name << " = " << res << endl;
return res;
}
Mat generateImage(const Size& sz, bool doBlur = true)
{
RNG rng;
Mat mat(sz, CV_8U);
mat = Scalar(0);
for(int y = 0; y < mat.rows; ++y)
for(int x = 0; x < mat.cols; ++x)
mat.at<uchar>(y, x) = (uchar)rng;
if (doBlur)
blur(mat, mat, Size(3, 3));
return mat;
}
Mat generateSample(const Size& sz)
{
Mat smpl(sz, CV_8U);
smpl = Scalar(0);
Point sc(smpl.cols/2, smpl.rows/2);
rectangle(smpl, Point(0,0), sc - Point(1,1), Scalar(255), CV_FILLED);
rectangle(smpl, sc, Point(smpl.cols, smpl.rows), Scalar(255), CV_FILLED);
return smpl;
}
bool CV_OptFlowTest::runDense(const Point& d)
{
Size matSize(40, 40);
Size movSize(8, 8);
Mat smpl = generateSample(movSize);
Mat prev = generateImage(matSize);
Mat curr = prev.clone();
Rect rect(Point(prev.cols/2, prev.rows/2) - Point(movSize.width/2, movSize.height/2), movSize);
Mat flowLK, flowBM, flowHS, flowFB, flowFB_G, flowBM_received, m1;
m1 = prev(rect); smpl.copyTo(m1);
m1 = curr(Rect(rect.tl() + d, rect.br() + d)); smpl.copyTo(m1);
calcOpticalFlowLK( prev, curr, Size(15, 15), flowLK);
calcOpticalFlowBM( prev, curr, Size(15, 15), Size(1, 1), Size(15, 15), 0, flowBM_received);
calcOpticalFlowHS( prev, curr, 0, 5, TermCriteria(TermCriteria::MAX_ITER, 400, 0), flowHS);
calcOpticalFlowFarneback( prev, curr, flowFB, 0.5, 3, std::max(d.x, d.y) + 10, 100, 6, 2, 0);
calcOpticalFlowFarneback( prev, curr, flowFB_G, 0.5, 3, std::max(d.x, d.y) + 10, 100, 6, 2, OPTFLOW_FARNEBACK_GAUSSIAN);
flowBM.create(prev.size(), CV_32FC2);
flowBM = Scalar(0);
Point origin((flowBM.cols - flowBM_received.cols)/2, (flowBM.rows - flowBM_received.rows)/2);
Mat wcp = flowBM(Rect(origin, flowBM_received.size()));
flowBM_received.copyTo(wcp);
double errorLK = showFlowAndCalcError("LK", prev, flowLK, rect, d);
double errorBM = showFlowAndCalcError("BM", prev, flowBM, rect, d);
double errorFB = showFlowAndCalcError("FB", prev, flowFB, rect, d);
double errorFBG = showFlowAndCalcError("FBG", prev, flowFB_G, rect, d);
double errorHS = showFlowAndCalcError("HS", prev, flowHS, rect, d); (void)errorHS;
//waitKey();
const double thres = 0.2;
if (errorLK > thres || errorBM > thres || errorFB > thres || errorFBG > thres /*|| errorHS > thres */)
{
ts->set_failed_test_info(cvtest::TS::FAIL_MISMATCH);
return false;
}
return true;
}
bool CV_OptFlowTest::runSparse()
{
Mat prev = imread(string(ts->get_data_path()) + "optflow/rock_1.bmp", 0);
Mat next = imread(string(ts->get_data_path()) + "optflow/rock_2.bmp", 0);
if (prev.empty() || next.empty())
{
ts->set_failed_test_info( cvtest::TS::FAIL_INVALID_TEST_DATA );
return false;
}
Mat cprev, cnext;
cvtColor(prev, cprev, CV_GRAY2BGR);
cvtColor(next, cnext, CV_GRAY2BGR);
vector<Point2f> prev_pts;
vector<Point2f> next_ptsOpt;
vector<Point2f> next_ptsAff;
vector<uchar> status_Opt;
vector<uchar> status_Aff;
vector<float> error;
vector<float> matrices;
Size netSize(10, 10);
Point2f center = Point(prev.cols/2, prev.rows/2);
for(int i = 0 ; i < netSize.width; ++i)
for(int j = 0 ; j < netSize.width; ++j)
{
Point2f p(i * float(prev.cols)/netSize.width, j * float(prev.rows)/netSize.height);
prev_pts.push_back((p - center) * 0.5f + center);
}
calcOpticalFlowPyrLK( prev, next, prev_pts, next_ptsOpt, status_Opt, error );
calcAffineFlowPyrLK ( prev, next, prev_pts, next_ptsAff, status_Aff, error, matrices);
const double expected_shift = 25;
const double thres = 1;
for(size_t i = 0; i < prev_pts.size(); ++i)
{
circle(cprev, prev_pts[i], 2, CV_RGB(255, 0, 0));
if (status_Opt[i])
{
circle(cnext, next_ptsOpt[i], 2, CV_RGB(0, 0, 255));
Point2f shift = prev_pts[i] - next_ptsOpt[i];
double n = sqrt(shift.ddot(shift));
if (fabs(n - expected_shift) > thres)
{
ts->set_failed_test_info(cvtest::TS::FAIL_MISMATCH);
return false;
}
}
if (status_Aff[i])
{
circle(cnext, next_ptsAff[i], 4, CV_RGB(0, 255, 0));
Point2f shift = prev_pts[i] - next_ptsAff[i];
double n = sqrt(shift.ddot(shift));
if (fabs(n - expected_shift) > thres)
{
ts->set_failed_test_info(cvtest::TS::FAIL_MISMATCH);
return false;
}
}
}
/*namedWindow("P"); imshow("P", cprev);
namedWindow("N"); imshow("N", cnext);
waitKey();*/
return true;
}
void CV_OptFlowTest::run( int /* start_from */)
{
if (!runDense(Point(3, 0)))
return;
if (!runDense(Point(0, 3)))
return;
//if (!runDense(Point(3, 3))) return; //probably LK works incorrectly in this case.
if (!runSparse())
return;
ts->set_failed_test_info(cvtest::TS::OK);
}
TEST(Legacy_OpticalFlow, accuracy) { CV_OptFlowTest test; test.safe_run(); }