Develop Reference

Detecting boxes via Hough Transform

Using HoughTransform I am trying to detect boxes and provide for a distinct color.
So far my understanding is a box is horizontal and vertical line .
my code is
lines = cv2.HoughLines(edges,1,np.pi/180, 50)
# The below for loop runs till r and theta values
# are in the range of the 2d array
for r,theta in lines[0]:
# Stores the value of cos(theta) in a
a = np.cos(theta)
# Stores the value of sin(theta) in b
b = np.sin(theta)
# x0 stores the value rcos(theta)
x0 = a*r
# y0 stores the value rsin(theta)
y0 = b*r
# x1 stores the rounded off value of (rcos(theta)-1000sin(theta))
x1 = int(x0 + 1000*(-b))
# y1 stores the rounded off value of (rsin(theta)+1000cos(theta))
y1 = int(y0 + 1000*(a))
# x2 stores the rounded off value of (rcos(theta)+1000sin(theta))
x2 = int(x0 - 1000*(-b))
# y2 stores the rounded off value of (rsin(theta)-1000cos(theta))
y2 = int(y0 - 1000*(a))
# cv2.line draws a line in img from the point(x1,y1) to (x2,y2).
# (255,255,255) denotes the colour of the line to be.
cv2.line(img,(x1,y1), (x2,y2), (255,255,255),2) `
What could i do so that the boxes can be colored or identified?

You should do vertical and horizontal line detection separately so that you can make them each more specific.
Go through all your lines and compile a list of intersections between horizontal and vertical line combinations
Now you have a list of 2d points that if you draw them should pretty much be on the corners of the boxes. The final step is to collect those points into meaningful sets.
To get those sets, I would start with the point nearest origin (just for the sake of starting somewhere). I would look through all the other points for the closest other point that has a greater x but is withing +-5 (or some configurable range) y of the starting point. Then do the same but in the y direction. You now have the bottom corner of the box. Which you could just complete and start your ocr on, but to be more robust, find the final corner as well.
Once all 4 corners are found, remove all of those points from your intersection array and add however you want to denote box locations into a new array. Rinse and repeat as now a different point will be nearest origin. Without actually testing this, I think it will choke (or need some conditional improvement for missing walls) on the K box but should be pretty generic to variable box shapes and sizes.
Edit 1: In testing, I am finding that it will probably be difficult to separate the close corners of two adjacent boxes. I think a more generic and robust solution would be to after you get collisions, do a point clustering operation at about 1/3 box min side length. This will average corners together with their nearest neighbor. So this will slightly change the strategy as you will need to use every corner twice (box to left and box to right) except for end points.
Wrote up some test code and it is functional, here are the outputs:
Code, sorry for c++ and not at all optimized, happy friday :)
//CPP libaries
#include <stdio.h>
#include <mutex>
#include <thread>
//Included libraries
//Note: these headers have to be before any opencv due to a namespace collision (could probably be fixed)
#include <opencv2/opencv.hpp>
#include "opencv2/imgproc/imgproc.hpp"
#include "opencv2/highgui/highgui.hpp"
using namespace cv;
// Finds the intersection of two lines, or returns false.
// The lines are defined by (o1, p1) and (o2, p2).
//https://stackoverflow.com/questions/7446126/opencv-2d-line-intersection-helper-function
bool intersection(Point2f o1, Point2f p1, Point2f o2, Point2f p2,
Point2f &r)
{
Point2f x = o2 - o1;
Point2f d1 = p1 - o1;
Point2f d2 = p2 - o2;
float cross = d1.x*d2.y - d1.y*d2.x;
if (abs(cross) < /*EPS*/1e-8)
return false;
double t1 = (x.x * d2.y - x.y * d2.x) / cross;
r = o1 + d1 * t1;
return true;
}
std::vector<Point2f> clusterPts(std::vector<Point2f> inputPts, double clusterRadius_Squared)
{
std::vector<Point2f> outputPts = std::vector<Point2f>();
while(inputPts.size()>0)
{
Point2f clusterCenter = inputPts[0];
while (true)
{
Point2f newClustCenter = Point2f(0, 0);
int averagingCount = 0;
std::vector<int> clusterIndicies = std::vector<int>();
for (int i = 0; i < inputPts.size(); i++)
{
if (clusterRadius_Squared >= pow(inputPts[i].x - clusterCenter.x, 2) + pow(inputPts[i].y - clusterCenter.y, 2))
{
newClustCenter.x += inputPts[i].x;
newClustCenter.y += inputPts[i].y;
averagingCount += 1;
clusterIndicies.push_back(i);
}
}
newClustCenter = newClustCenter / (double)averagingCount;
if (newClustCenter == clusterCenter)
{
//remove all points inside cluster from inputPts, stash cluster center, and break inner while loop
std::vector<Point2f> remainingPts = std::vector<Point2f>();
for (int i = 0; i < inputPts.size(); i++)
{
if (std::find(clusterIndicies.begin(), clusterIndicies.end(), i) == clusterIndicies.end())
{
remainingPts.push_back(inputPts[i]);
}
}
inputPts = remainingPts;
outputPts.push_back(clusterCenter);
break;
}
else
{
clusterCenter = newClustCenter;
}
}
}
return outputPts;
}
std::vector<Rect> findBoxes(std::vector<Point2f> corners, bool shrinkBoxes = false, int boxSideLength_Guess = 50)
{
std::vector<Rect> outBoxes = std::vector<Rect>();
int approxBoxSize = 1000 * boxSideLength_Guess;
while (corners.size()>4)
{
//find point above or below (these points will be removed from array after used)
int secondPtIndex = -1;
for (int i = 1; i < corners.size(); i++)
{
if (abs(corners[i].x - corners[0].x) < boxSideLength_Guess / 2.0)
{
secondPtIndex = i;
break;
}
}
if (secondPtIndex == -1)
{
std::cout << "bad box point tossed" << std::endl;
corners.erase(corners.begin() + 0);
continue;
}
//now search for closest same level point on either side
int thirdIndexRight = -1;
int thirdIndexLeft = -1;
double minDistRight = approxBoxSize;
double minDistLeft = -approxBoxSize;
for (int i = 2; i < corners.size(); i++)
{
if (abs(corners[i].y - corners[secondPtIndex].y) < boxSideLength_Guess / 2.0)
{
double dist = corners[i].x - corners[secondPtIndex].x;
if (dist < 0 && dist > minDistLeft) //check left
{
minDistLeft = dist;
thirdIndexLeft = i;
}
else if(dist > 0 && dist < minDistRight) //check right
{
minDistRight = dist;
thirdIndexRight = i;
}
}
}
if (thirdIndexLeft != -1) { approxBoxSize = 1.5 * abs(minDistLeft); }
if (thirdIndexRight != -1) { approxBoxSize = 1.5 * minDistRight; }
int fourthIndexRight = -1;
int fourthIndexLeft = -1;
for (int i = 1; i < corners.size(); i++)
{
if (i == thirdIndexLeft || i == thirdIndexRight) { continue; }
if (thirdIndexLeft != -1 && abs(corners[i].x - corners[thirdIndexLeft].x) < boxSideLength_Guess / 2.0)
{ fourthIndexLeft = i; }
if (thirdIndexRight != -1 && abs(corners[i].x - corners[thirdIndexRight].x) < boxSideLength_Guess / 2.0)
{ fourthIndexRight = i; }
}
if (!shrinkBoxes)
{
if (fourthIndexRight != -1)
{
outBoxes.push_back(Rect(corners[0], corners[thirdIndexRight]));
}
if (fourthIndexLeft != -1)
{
outBoxes.push_back(Rect(corners[0], corners[thirdIndexLeft]));
}
}
else
{
if (fourthIndexRight != -1)
{
outBoxes.push_back(Rect(corners[0] * 0.90 + corners[thirdIndexRight] *0.10, corners[0] * 0.10 + corners[thirdIndexRight] * 0.90));
}
if (fourthIndexLeft != -1)
{
outBoxes.push_back(Rect(corners[0] * 0.90 + corners[thirdIndexLeft] * 0.10, corners[0] * 0.10 + corners[thirdIndexLeft] * 0.90));
}
}
corners.erase(corners.begin() + secondPtIndex);
corners.erase(corners.begin() + 0);
}
std::cout << approxBoxSize << std::endl;
return outBoxes;
}
int main(int argc, char** argv)
{
Mat image = imread("../../resources/images/boxPic.png", CV_LOAD_IMAGE_GRAYSCALE);
imshow("source", image);
//namedWindow("Display window", WINDOW_AUTOSIZE);// Create a window for display.
//imshow("Display window", image); // Show our image inside it.
Mat edges, lineOverlay, cornerOverlay, finalBoxes;
Canny(image, edges, 50, 200, 3);
//edges = image;
//cvtColor(image, edges, COLOR_GRAY2BGR);
cvtColor(image, lineOverlay, COLOR_GRAY2BGR);
cvtColor(image, cornerOverlay, COLOR_GRAY2BGR);
cvtColor(image, finalBoxes, COLOR_GRAY2BGR);
std::cout << image.cols << " , "<<image.rows << std::endl;
std::vector<Vec2f> linesHorizontal;
std::vector<Point> ptsLH;
HoughLines(edges, linesHorizontal, 5, CV_PI / 180, 2 * edges.cols * 0.6, 0.0,0.0, CV_PI / 4, 3 * CV_PI / 4);
std::vector<Vec2f> linesVertical;
std::vector<Point> ptsLV;
HoughLines(edges, linesVertical, 5, CV_PI / 180, 2 * edges.rows * 0.6,0,0,-CV_PI/32,CV_PI/32);
for (size_t i = 0; i < linesHorizontal.size(); i++)
{
float rho = linesHorizontal[i][0], theta = linesHorizontal[i][1];
Point pt1, pt2;
double a = cos(theta), b = sin(theta);
double x0 = a * rho, y0 = b * rho;
pt1.x = cvRound(x0 + 1000 * (-b));
pt1.y = cvRound(y0 + 1000 * (a));
pt2.x = cvRound(x0 - 1000 * (-b));
pt2.y = cvRound(y0 - 1000 * (a));
ptsLH.push_back(pt1);
ptsLH.push_back(pt2);
line(lineOverlay, pt1, pt2, Scalar(0, 0, 255), 1, LINE_AA);
}
for (size_t i = 0; i < linesVertical.size(); i++)
{
float rho = linesVertical[i][0], theta = linesVertical[i][1];
Point pt1, pt2;
double a = cos(theta), b = sin(theta);
double x0 = a * rho, y0 = b * rho;
pt1.x = cvRound(x0 + 1000 * (-b));
pt1.y = cvRound(y0 + 1000 * (a));
pt2.x = cvRound(x0 - 1000 * (-b));
pt2.y = cvRound(y0 - 1000 * (a));
ptsLV.push_back(pt1);
ptsLV.push_back(pt2);
line(lineOverlay, pt1, pt2, Scalar(0, 255, 0), 1, LINE_AA);
}
imshow("edged", edges);
imshow("detected lines", lineOverlay);
//look for collisions
std::vector<Point2f> xPts;
for (size_t i = 0; i < linesHorizontal.size(); i++)
{
for (size_t ii = 0; ii < linesVertical.size(); ii++)
{
Point2f xPt;
bool intersectionExists = intersection(ptsLH[2 * i], ptsLH[2 * i + 1], ptsLV[2 * ii], ptsLV[2 * ii + 1], xPt);
if (intersectionExists)
{
xPts.push_back(xPt);
}
}
}
waitKey(1000);
std::vector<Point2f> boxCorners = clusterPts(xPts, 25*25);
for (int i = 0; i < boxCorners.size(); i++)
{
circle(cornerOverlay, boxCorners[i], 5, Scalar(0, 255, 0), 2);
}
imshow("detected corners", cornerOverlay);
//group make boxes for groups of points
std::vector<Rect> ocrBoxes = findBoxes(boxCorners,true);
for (int i = 0; i < ocrBoxes.size(); i++)
{
if (i % 3 == 0) { rectangle(finalBoxes, ocrBoxes[i], Scalar(255, 0, 0), 2); }
else if(i % 3 == 1) { rectangle(finalBoxes, ocrBoxes[i], Scalar(0, 255, 0), 2); }
else if (i % 3 == 2) { rectangle(finalBoxes, ocrBoxes[i], Scalar(0, 0, 255), 2); }
}
imshow("detected boxes", finalBoxes);
waitKey(0); // Wait for a keystroke in the window
return 0;
}

cv2.estimateRigidTransform minimal number of points?

What is minimal number of points needed for cv2.estimateRigidTransform?
As I understand with fullAffine=False it have 4 degrees of freedom, so 2 points should be sufficient.
However:
Using 2 numpy array as input:
src_pts_subset.shape (2, 2)
tgt_pts_subset.shape (2, 2)
type(src_pts_subset) <class 'numpy.ndarray'>
type(tgt_pts_subset) <class 'numpy.ndarray'>
src_pts_subset.dtype int64
tgt_pts_subset.dtype int64
To
m = cv2.estimateRigidTransform(src_pts, tgt_pts, fullAffine=False)
Gives me None.

Theoretically only 2 pair points is needed for non-full affine setting as explained thoroughly by ngia ho. However, upon inspection to the source code in openCV, a minimum of 3 pair points is required for the function to return a value calculated by RANSAC.
I have included the corresponding function below for your reference. It is located in the file lkpyramid.cpp in the openCV source file.
cv::Mat cv::estimateRigidTransform( InputArray src1, InputArray src2, bool fullAffine )
{
return estimateRigidTransform(src1, src2, fullAffine, 500, 0.5, 3);
}
cv::Mat cv::estimateRigidTransform( InputArray src1, InputArray src2, bool fullAffine, int ransacMaxIters, double ransacGoodRatio,
const int ransacSize0)
{
CV_INSTRUMENT_REGION()
Mat M(2, 3, CV_64F), A = src1.getMat(), B = src2.getMat();
const int COUNT = 15;
const int WIDTH = 160, HEIGHT = 120;
std::vector<Point2f> pA, pB;
std::vector<int> good_idx;
std::vector<uchar> status;
double scale = 1.;
int i, j, k, k1;
RNG rng((uint64)-1);
int good_count = 0;
if( ransacSize0 < 3 )
CV_Error( Error::StsBadArg, "ransacSize0 should have value bigger than 2.");
if( ransacGoodRatio > 1 || ransacGoodRatio < 0)
CV_Error( Error::StsBadArg, "ransacGoodRatio should have value between 0 and 1");
if( A.size() != B.size() )
CV_Error( Error::StsUnmatchedSizes, "Both input images must have the same size" );
if( A.type() != B.type() )
CV_Error( Error::StsUnmatchedFormats, "Both input images must have the same data type" );
int count = A.checkVector(2);
if( count > 0 )
{
A.reshape(2, count).convertTo(pA, CV_32F);
B.reshape(2, count).convertTo(pB, CV_32F);
}
else if( A.depth() == CV_8U )
{
int cn = A.channels();
CV_Assert( cn == 1 || cn == 3 || cn == 4 );
Size sz0 = A.size();
Size sz1(WIDTH, HEIGHT);
scale = std::max(1., std::max( (double)sz1.width/sz0.width, (double)sz1.height/sz0.height ));
sz1.width = cvRound( sz0.width * scale );
sz1.height = cvRound( sz0.height * scale );
bool equalSizes = sz1.width == sz0.width && sz1.height == sz0.height;
if( !equalSizes || cn != 1 )
{
Mat sA, sB;
if( cn != 1 )
{
Mat gray;
cvtColor(A, gray, COLOR_BGR2GRAY);
resize(gray, sA, sz1, 0., 0., INTER_AREA);
cvtColor(B, gray, COLOR_BGR2GRAY);
resize(gray, sB, sz1, 0., 0., INTER_AREA);
}
else
{
resize(A, sA, sz1, 0., 0., INTER_AREA);
resize(B, sB, sz1, 0., 0., INTER_AREA);
}
A = sA;
B = sB;
}
int count_y = COUNT;
int count_x = cvRound((double)COUNT*sz1.width/sz1.height);
count = count_x * count_y;
pA.resize(count);
pB.resize(count);
status.resize(count);
for( i = 0, k = 0; i < count_y; i++ )
for( j = 0; j < count_x; j++, k++ )
{
pA[k].x = (j+0.5f)*sz1.width/count_x;
pA[k].y = (i+0.5f)*sz1.height/count_y;
}
// find the corresponding points in B
calcOpticalFlowPyrLK(A, B, pA, pB, status, noArray(), Size(21, 21), 3,
TermCriteria(TermCriteria::MAX_ITER,40,0.1));
// repack the remained points
for( i = 0, k = 0; i < count; i++ )
if( status[i] )
{
if( i > k )
{
pA[k] = pA[i];
pB[k] = pB[i];
}
k++;
}
count = k;
pA.resize(count);
pB.resize(count);
}
else
CV_Error( Error::StsUnsupportedFormat, "Both input images must have either 8uC1 or 8uC3 type" );
good_idx.resize(count);
if( count < ransacSize0 )
return Mat();
Rect brect = boundingRect(pB);
std::vector<Point2f> a(ransacSize0);
std::vector<Point2f> b(ransacSize0);
// RANSAC stuff:
// 1. find the consensus
for( k = 0; k < ransacMaxIters; k++ )
{
std::vector<int> idx(ransacSize0);
// choose random 3 non-complanar points from A & B
for( i = 0; i < ransacSize0; i++ )
{
for( k1 = 0; k1 < ransacMaxIters; k1++ )
{
idx[i] = rng.uniform(0, count);
for( j = 0; j < i; j++ )
{
if( idx[j] == idx[i] )
break;
// check that the points are not very close one each other
if( fabs(pA[idx[i]].x - pA[idx[j]].x) +
fabs(pA[idx[i]].y - pA[idx[j]].y) < FLT_EPSILON )
break;
if( fabs(pB[idx[i]].x - pB[idx[j]].x) +
fabs(pB[idx[i]].y - pB[idx[j]].y) < FLT_EPSILON )
break;
}
if( j < i )
continue;
if( i+1 == ransacSize0 )
{
// additional check for non-complanar vectors
a[0] = pA[idx[0]];
a[1] = pA[idx[1]];
a[2] = pA[idx[2]];
b[0] = pB[idx[0]];
b[1] = pB[idx[1]];
b[2] = pB[idx[2]];
double dax1 = a[1].x - a[0].x, day1 = a[1].y - a[0].y;
double dax2 = a[2].x - a[0].x, day2 = a[2].y - a[0].y;
double dbx1 = b[1].x - b[0].x, dby1 = b[1].y - b[0].y;
double dbx2 = b[2].x - b[0].x, dby2 = b[2].y - b[0].y;
const double eps = 0.01;
if( fabs(dax1*day2 - day1*dax2) < eps*std::sqrt(dax1*dax1+day1*day1)*std::sqrt(dax2*dax2+day2*day2) ||
fabs(dbx1*dby2 - dby1*dbx2) < eps*std::sqrt(dbx1*dbx1+dby1*dby1)*std::sqrt(dbx2*dbx2+dby2*dby2) )
continue;
}
break;
}
if( k1 >= ransacMaxIters )
break;
}
if( i < ransacSize0 )
continue;
// estimate the transformation using 3 points
getRTMatrix( a, b, 3, M, fullAffine );
const double* m = M.ptr<double>();
for( i = 0, good_count = 0; i < count; i++ )
{
if( std::abs( m[0]*pA[i].x + m[1]*pA[i].y + m[2] - pB[i].x ) +
std::abs( m[3]*pA[i].x + m[4]*pA[i].y + m[5] - pB[i].y ) < std::max(brect.width,brect.height)*0.05 )
good_idx[good_count++] = i;
}
if( good_count >= count*ransacGoodRatio )
break;
}
if( k >= ransacMaxIters )
return Mat();
if( good_count < count )
{
for( i = 0; i < good_count; i++ )
{
j = good_idx[i];
pA[i] = pA[j];
pB[i] = pB[j];
}
}
getRTMatrix( pA, pB, good_count, M, fullAffine );
M.at<double>(0, 2) /= scale;
M.at<double>(1, 2) /= scale;
return M;
}

How to check if an image is flipped or not with respect to base image

In OpenCV I can flip the image horizontally and vertically using this code :
import cv2
img=cv2.imread('1.png')
rimg=img.copy()
fimg=img.copy()
rimg=cv2.flip(img,1)
fimg=cv2.flip(img,0)
cv2.imshow("Original", img)
cv2.imshow("vertical flip", rimg)
cv2.imshow("horizontal flip", fimg)
cv2.waitKey(0)
cv2.destroyAllWindows()
My question is: is there any way to check that the input image is flipped with respect to the base image? I just want to know the steps to check if the image is flipped or not.

Here is a pseudo c++ solution:
bool isSameSize(Image a, Image b)
{
return a.width() == b.width() && a.height() == b.height();
}
bool isNotFlipped(Image a, Image b)
{
if(!isSameSize(a, b)) return false;
for(int y = 0; y < a.height(); ++y)
for(int x = 0; x < a.width(); ++x)
{
if(a.pixel(x, y) != b.pixel(x, y))
return false;
}
return true;
}
bool isFlippedHorizontal(Image a, Image b)
{
if(!isSameSize(a, b)) return false;
int w = a.width();
for(int y = 0; y < a.height(); ++y)
for(int x = 0; x < a.width(); ++x)
{
if(a.pixel(x, y) != b.pixel(w - x - 1, y))
return false;
}
return true;
}
bool isFlippedVertical(Image a, Image b)
{
if(!isSameSize(a, b)) return false;
int h = a.height();
for(int y = 0; y < a.height(); ++y)
for(int x = 0; x < a.width(); ++x)
{
if(a.pixel(x, y) != b.pixel(x, h - y - 1))
return false;
}
return true;
}
bool isFlippedBoth(Image a, Image b)
{
if(!isSameSize(a, b)) return false;
int w = a.width();
int h = a.height();
for(int y = 0; y < a.height(); ++y)
for(int x = 0; x < a.width(); ++x)
{
if(a.pixel(x, y) != b.pixel(w - x - 1, h - y - 1))
return false;
}
return true;
}
HTH

In general, no. If the image is not symmetric in some fashion, then you can probably just flip along different axes and do a quick comparison to check if any are non-equal.
def is_horizontally_flipped(orig, img):
tst = cv2.flip(img, 1)
return np.any(orig != tst)

optical flow .flo files

I have a few questions for doing optical flow projects. I use Python 2 (planning to use lasagne to use deep learning to learn optical flow), and don't know how to convert the c++ functions to that of python in visualization of the flows.
I downloaded (from http://vision.middlebury.edu/flow/data/comp/zip/other-gt-flow.zip) some image pairs where I have to estimate their optical flow, and their ground truth flow (.flo file). The problem is, when I read the .flo file into the program, it is a vectorized code. How do I view them like how they show in the webpage (http://vision.middlebury.edu/flow/data/)? I read from various sources and tried the following, but doesn't work.
In evaluating EPE (end point error) in what form should I have my prediction to be compared with the .flo file?
The code:
################################ Reading flow file ################################
f = open('flow10.flo', 'rb')
x = np.fromfile(f, np.int32, count=1) # not sure what this gives
w = np.fromfile(f, np.int32, count=1) # width
h = np.fromfile(f, np.int32, count=1) # height
print 'x %d, w %d, h %d flo file' % (x, w, h)
data = np.fromfile(f, np.float32) # vector
data_2D = np.reshape(data, newshape=(388,584,2)); # convert to x,y - flow
x = data_2D[...,0]; y = data_2D[...,1];
################################ visualising flow file ################################
mag, ang = cv2.cartToPolar(x,y)
hsv = np.zeros_like(x)
hsv = np.array([ hsv,hsv,hsv ])
hsv = np.reshape(hsv, (388,584,3)); # having rgb channel
hsv[...,1] = 255; # full green channel
hsv[...,0] = ang*180/np.pi/2 # angle in pi
hsv[...,2] = cv2.normalize(mag,None,0,255,cv2.NORM_MINMAX) # magnitude [0,255]
bgr = cv2.cvtColor(hsv,cv2.COLOR_HSV2BGR)
bgr = draw_hsv(data_2D)
cv2.imwrite('opticalhsv.png',bgr)

On Middlebury's page there is a zip file called flow-code (http://vision.middlebury.edu/flow/code/flow-code.zip), which provides a tool called color_flow to convert those .flo files to color images.
On the other hand, if you want to implement your own code to do the transformation, i have this piece of code (i cannot provide the original author, it has been some time) that helps you to first compute the color:
static Vec3b computeColor(float fx, float fy)
{
static bool first = true;
// relative lengths of color transitions:
// these are chosen based on perceptual similarity
// (e.g. one can distinguish more shades between red and yellow
// than between yellow and green)
const int RY = 15;
const int YG = 6;
const int GC = 4;
const int CB = 11;
const int BM = 13;
const int MR = 6;
const int NCOLS = RY + YG + GC + CB + BM + MR;
static Vec3i colorWheel[NCOLS];
if (first)
{
int k = 0;
for (int i = 0; i < RY; ++i, ++k)
colorWheel[k] = Vec3i(255, 255 * i / RY, 0);
for (int i = 0; i < YG; ++i, ++k)
colorWheel[k] = Vec3i(255 - 255 * i / YG, 255, 0);
for (int i = 0; i < GC; ++i, ++k)
colorWheel[k] = Vec3i(0, 255, 255 * i / GC);
for (int i = 0; i < CB; ++i, ++k)
colorWheel[k] = Vec3i(0, 255 - 255 * i / CB, 255);
for (int i = 0; i < BM; ++i, ++k)
colorWheel[k] = Vec3i(255 * i / BM, 0, 255);
for (int i = 0; i < MR; ++i, ++k)
colorWheel[k] = Vec3i(255, 0, 255 - 255 * i / MR);
first = false;
}
const float rad = sqrt(fx * fx + fy * fy);
const float a = atan2(-fy, -fx) / (float)CV_PI;
const float fk = (a + 1.0f) / 2.0f * (NCOLS - 1);
const int k0 = static_cast<int>(fk);
const int k1 = (k0 + 1) % NCOLS;
const float f = fk - k0;
Vec3b pix;
for (int b = 0; b < 3; b++)
{
const float col0 = colorWheel[k0][b] / 255.f;
const float col1 = colorWheel[k1][b] / 255.f;
float col = (1 - f) * col0 + f * col1;
if (rad <= 1)
col = 1 - rad * (1 - col); // increase saturation with radius
else
col *= .75; // out of range
pix[2 - b] = static_cast<uchar>(255.f * col);
}
return pix;
}
Then it calls the above function for all the pixels:
static void drawOpticalFlow(const Mat_<Point2f>& flow, Mat& dst, float maxmotion = -1)
{
dst.create(flow.size(), CV_8UC3);
dst.setTo(Scalar::all(0));
// determine motion range:
float maxrad = maxmotion;
if (maxmotion <= 0)
{
maxrad = 1;
for (int y = 0; y < flow.rows; ++y)
{
for (int x = 0; x < flow.cols; ++x)
{
Point2f u = flow(y, x);
if (!isFlowCorrect(u))
continue;
maxrad = max(maxrad, sqrt(u.x * u.x + u.y * u.y));
}
}
}
for (int y = 0; y < flow.rows; ++y)
{
for (int x = 0; x < flow.cols; ++x)
{
Point2f u = flow(y, x);
if (isFlowCorrect(u))
dst.at<Vec3b>(y, x) = computeColor(u.x / maxrad, u.y / maxrad);
}
}
}
This is for my use in OpenCV, but the code help should anyone who wants achieve something similar.

OpenCV how to smooth contour, reducing noise

I extracted the contours of an image, that you can see here:
However, it has some noise.
How can I smooth the noise? I did a close up to make clearer what I want to meant
Original image that I've used:
Code:
rMaskgray = cv2.imread('redmask.jpg', cv2.CV_LOAD_IMAGE_GRAYSCALE)
(thresh, binRed) = cv2.threshold(rMaskgray, 50, 255, cv2.THRESH_BINARY)
Rcontours, hier_r = cv2.findContours(binRed,cv2.RETR_CCOMP,cv2.CHAIN_APPROX_SIMPLE)
r_areas = [cv2.contourArea(c) for c in Rcontours]
max_rarea = np.max(r_areas)
CntExternalMask = np.ones(binRed.shape[:2], dtype="uint8") * 255
for c in Rcontours:
if(( cv2.contourArea(c) > max_rarea * 0.70) and (cv2.contourArea(c)< max_rarea)):
cv2.drawContours(CntExternalMask,[c],-1,0,1)
cv2.imwrite('contour1.jpg', CntExternalMask)

Try an upgrade to OpenCV 3.1.0. After some code adaptations for the new version as shown below, I tried it out with OpenCV version 3.1.0 and did not see any of the effects you are describing.
import cv2
import numpy as np
print cv2.__version__
rMaskgray = cv2.imread('5evOn.jpg', 0)
(thresh, binRed) = cv2.threshold(rMaskgray, 50, 255, cv2.THRESH_BINARY)
_, Rcontours, hier_r = cv2.findContours(binRed,cv2.RETR_CCOMP,cv2.CHAIN_APPROX_SIMPLE)
r_areas = [cv2.contourArea(c) for c in Rcontours]
max_rarea = np.max(r_areas)
CntExternalMask = np.ones(binRed.shape[:2], dtype="uint8") * 255
for c in Rcontours:
if(( cv2.contourArea(c) > max_rarea * 0.70) and (cv2.contourArea(c)< max_rarea)):
cv2.drawContours(CntExternalMask,[c],-1,0,1)
cv2.imwrite('contour1.jpg', CntExternalMask)

I don't know if is it ok to provide Java code - but I implemented Gaussian smoothing for openCV contour. Logic and theory is taken from here https://www.morethantechnical.com/2012/12/07/resampling-smoothing-and-interest-points-of-curves-via-css-in-opencv-w-code/
package CurveTools;
import org.apache.log4j.Logger;
import org.opencv.core.Mat;
import org.opencv.core.MatOfPoint;
import org.opencv.core.Point;
import java.util.ArrayList;
import java.util.List;
import static org.opencv.core.CvType.CV_64F;
import static org.opencv.imgproc.Imgproc.getGaussianKernel;
class CurveSmoother {
private double[] g, dg, d2g, gx, dx, d2x;
private double gx1, dgx1, d2gx1;
public double[] kappa, smoothX, smoothY;
public double[] contourX, contourY;
/* 1st and 2nd derivative of 1D gaussian */
void getGaussianDerivs(double sigma, int M) {
int L = (M - 1) / 2;
double sigma_sq = sigma * sigma;
double sigma_quad = sigma_sq * sigma_sq;
dg = new double[M];
d2g = new double[M];
g = new double[M];
Mat tmpG = getGaussianKernel(M, sigma, CV_64F);
for (double i = -L; i < L + 1.0; i += 1.0) {
int idx = (int) (i + L);
g[idx] = tmpG.get(idx, 0)[0];
// from http://www.cedar.buffalo.edu/~srihari/CSE555/Normal2.pdf
dg[idx] = -i * g[idx] / sigma_sq;
d2g[idx] = (-sigma_sq + i * i) * g[idx] / sigma_quad;
}
}
/* 1st and 2nd derivative of smoothed curve point */
void getdX(double[] x, int n, double sigma, boolean isOpen) {
int L = (g.length - 1) / 2;
gx1 = dgx1 = d2gx1 = 0.0;
for (int k = -L; k < L + 1; k++) {
double x_n_k;
if (n - k < 0) {
if (isOpen) {
//open curve - mirror values on border
x_n_k = x[-(n - k)];
} else {
//closed curve - take values from end of curve
x_n_k = x[x.length + (n - k)];
}
} else if (n - k > x.length - 1) {
if (isOpen) {
//mirror value on border
x_n_k = x[n + k];
} else {
x_n_k = x[(n - k) - x.length];
}
} else {
x_n_k = x[n - k];
}
gx1 += x_n_k * g[k + L]; //gaussians go [0 -> M-1]
dgx1 += x_n_k * dg[k + L];
d2gx1 += x_n_k * d2g[k + L];
}
}
/* 0th, 1st and 2nd derivatives of whole smoothed curve */
void getdXcurve(double[] x, double sigma, boolean isOpen) {
gx = new double[x.length];
dx = new double[x.length];
d2x = new double[x.length];
for (int i = 0; i < x.length; i++) {
getdX(x, i, sigma, isOpen);
gx[i] = gx1;
dx[i] = dgx1;
d2x[i] = d2gx1;
}
}
/*
compute curvature of curve after gaussian smoothing
from "Shape similarity retrieval under affine transforms", Mokhtarian & Abbasi 2002
curvex - x position of points
curvey - y position of points
kappa - curvature coeff for each point
sigma - gaussian sigma
*/
void computeCurveCSS(double[] curvex, double[] curvey, double sigma, boolean isOpen) {
int M = (int) Math.round((10.0 * sigma + 1.0) / 2.0) * 2 - 1;
assert (M % 2 == 1); //M is an odd number
getGaussianDerivs(sigma, M);//, g, dg, d2g
double[] X, XX, Y, YY;
getdXcurve(curvex, sigma, isOpen);
smoothX = gx.clone();
X = dx.clone();
XX = d2x.clone();
getdXcurve(curvey, sigma, isOpen);
smoothY = gx.clone();
Y = dx.clone();
YY = d2x.clone();
kappa = new double[curvex.length];
for (int i = 0; i < curvex.length; i++) {
// Mokhtarian 02' eqn (4)
kappa[i] = (X[i] * YY[i] - XX[i] * Y[i]) / Math.pow(X[i] * X[i] + Y[i] * Y[i], 1.5);
}
}
/* find zero crossings on curvature */
ArrayList<Integer> findCSSInterestPoints() {
assert (kappa != null);
ArrayList<Integer> crossings = new ArrayList<>();
for (int i = 0; i < kappa.length - 1; i++) {
if ((kappa[i] < 0.0 && kappa[i + 1] > 0.0) || kappa[i] > 0.0 && kappa[i + 1] < 0.0) {
crossings.add(i);
}
}
return crossings;
}
public void polyLineSplit(MatOfPoint pl) {
contourX = new double[pl.height()];
contourY = new double[pl.height()];
for (int j = 0; j < contourX.length; j++) {
contourX[j] = pl.get(j, 0)[0];
contourY[j] = pl.get(j, 0)[1];
}
}
public MatOfPoint polyLineMerge(double[] xContour, double[] yContour) {
assert (xContour.length == yContour.length);
MatOfPoint pl = new MatOfPoint();
List<Point> list = new ArrayList<>();
for (int j = 0; j < xContour.length; j++)
list.add(new Point(xContour[j], yContour[j]));
pl.fromList(list);
return pl;
}
MatOfPoint smoothCurve(MatOfPoint curve, double sigma) {
int M = (int) Math.round((10.0 * sigma + 1.0) / 2.0) * 2 - 1;
assert (M % 2 == 1); //M is an odd number
//create kernels
getGaussianDerivs(sigma, M);
polyLineSplit(curve);
getdXcurve(contourX, sigma, false);
smoothX = gx.clone();
getdXcurve(contourY, sigma, false);
smoothY = gx;
Logger.getRootLogger().info("Smooth curve len: " + smoothX.length);
return polyLineMerge(smoothX, smoothY);
}
}