Ask Your Question

Revision history [back]

click to hide/show revision 1
initial version

Blurred QRcode detection from video stream with OpenCV

I'm trying to extrapolate from a video stream, some QR codes that have been modified to not be easily recognizable. I've to use OpenCV. The problem is that the code is able to recognize only the qr codes well visible, while the qr codes that it must recognize are blurred. I believe that the problem is some parameters, but i can't find it. According to you where is the problem?

this is the code i use: https://stackoverflow.com/questions/51143934/blurred-qrcode-detection-from-video-stream-with-opencv#comment89284616_51143934

Blurred QRcode detection from video stream with OpenCV

I'm trying to extrapolate from a video stream, some QR codes that have been modified to not be easily recognizable. I've to use OpenCV. The problem is that the code is able to recognize only the qr codes well visible, while the qr codes that it must recognize are blurred. I believe that the problem is some parameters, but i can't find it. According to you where is the problem?

this is the code i use: https://stackoverflow.com/questions/51143934/blurred-qrcode-detection-from-video-stream-with-opencv#comment89284616_51143934

#include <opencv2/opencv.hpp>
#include <iostream>
#include <cmath>

using namespace cv;
using namespace std;

const int CV_QR_NORTH = 0;
const int CV_QR_EAST = 1;
const int CV_QR_SOUTH = 2;
const int CV_QR_WEST = 3;

int counter = 0;



float cv_distance(Point2f P, Point2f Q);                    // Get Distance between two points
float cv_lineEquation(Point2f L, Point2f M, Point2f J);     // Perpendicular Distance of a Point J from line formed by Points L and M; Solution to equation of the line Val = ax+by+c 
float cv_lineSlope(Point2f L, Point2f M, int& alignement);  // Slope of a line by two Points L and M on it; Slope of line, S = (x1 -x2) / (y1- y2)
void cv_getVertices(vector<vector<Point> > contours, int c_id, float slope, vector<Point2f>& X);
void cv_updateCorner(Point2f P, Point2f ref, float& baseline, Point2f& corner);
void cv_updateCornerOr(int orientation, vector<Point2f> IN, vector<Point2f> &OUT);
bool getIntersectionPoint(Point2f a1, Point2f a2, Point2f b1, Point2f b2, Point2f& intersection);
float cross(Point2f v1, Point2f v2);

// Start of Main Loop
//------------------------------------------------------------------------------------------------------------------------
int main(int argc, char **argv)
{
    VideoCapture capture;
    capture.open("video.mp4");

//Mat image = imread(argv[1]);
Mat image;

if (!capture.isOpened()) {
    cerr << " ERR: Unable find input Video source." << endl;
    return -1;
}

//Step  : Capture a frame from Image Input for creating and initializing manipulation variables
//Info  : Inbuilt functions from OpenCV
//Note  : 

capture >> image;
if (image.empty()) {
    cerr << "ERR: Unable to query image from capture device.\n" << endl;
    return -1;
}


// Creation of Intermediate 'Image' Objects required later
Mat gray(image.size(), CV_MAKETYPE(image.depth(), 1));          // To hold Grayscale Image
Mat edges(image.size(), CV_MAKETYPE(image.depth(), 1));         // To hold Grayscale Image
Mat traces(image.size(), CV_8UC3);                              // For Debug Visuals
Mat qr, qr_raw, qr_gray, qr_thres;


vector<vector<Point> > contours;
vector<Vec4i> hierarchy;
vector<Point> pointsseq;    //used to save the approximated sides of each contour

int mark, A, B, C, top, right, bottom, median1, median2, outlier;
float AB, BC, CA, dist, slope, areat, arear, areab, large, padding;

int align, orientation;

int DBG = 1;                        // Debug Flag

int key = 0;
while (key != 'q')              // While loop to query for Image Input frame
{

    traces = Scalar(0, 0, 0);

    qr_raw = Mat::zeros(100, 100, CV_8UC3); 
    qr = Mat::zeros(100, 100, CV_8UC3);
    qr_gray = Mat::zeros(100, 100, CV_8UC1);
    qr_thres = Mat::zeros(100, 100, CV_8UC1);

    capture >> image;                       // Capture Image from Image Input

    cvtColor(image, gray, CV_RGB2GRAY);     // Convert Image captured from Image Input to GrayScale 
    Canny(gray, edges, 100, 200, 3);        // Apply Canny edge detection on the gray image


    findContours(edges, contours, hierarchy, RETR_TREE, CHAIN_APPROX_SIMPLE); 

    mark = 0;                               // Reset all detected marker count for this frame

                                            // Get Moments for all Contours and the mass centers
    vector<Moments> mu(contours.size());
    vector<Point2f> mc(contours.size());

    for (int i = 0; i < contours.size(); i++)
    {
        mu[i] = moments(contours[i], false);
        mc[i] = Point2f(mu[i].m10 / mu[i].m00, mu[i].m01 / mu[i].m00);
    }


    // Start processing the contour data

    // Find Three repeatedly enclosed contours A,B,C
    // NOTE: 1. Contour enclosing other contours is assumed to be the three Alignment markings of the QR code.
    // 2. Alternately, the Ratio of areas of the "concentric" squares can also be used for identifying base Alignment markers.
    // The below demonstrates the first method

    for (int i = 0; i < contours.size(); i++)
    {
        //Find the approximated polygon of the contour we are examining
        approxPolyDP(contours[i], pointsseq, arcLength(contours[i], true)*0.02, true); 
        if (pointsseq.size() == 4)      // only quadrilaterals contours are examined
        {
            int k = i;
            int c = 0;


            while (hierarchy[k][2] != -1) 
            {
                k = hierarchy[k][2]; 
                c = c + 1;
            }
            if (hierarchy[k][2] != -1) 
                c = c + 1;

            if (c >= 5) 
            {
                if (mark == 0)      A = i; 
                else if (mark == 1) B = i;      // i.e., A is already found, assign current contour to B
                else if (mark == 2) C = i;      // i.e., A and B are already found, assign current contour to C
                mark = mark + 1;
            }
        }
    }


    if (mark >= 3)      // Ensure we have (atleast 3; namely A,B,C) 'Alignment Markers' discovered
    {
        // We have found the 3 markers for the QR code; Now we need to determine which of them are 'top', 'right' and 'bottom' markers

        // Determining the 'top' marker
        // Vertex of the triangle NOT involved in the longest side is the 'outlier'

        AB = cv_distance(mc[A], mc[B]);
        BC = cv_distance(mc[B], mc[C]);
        CA = cv_distance(mc[C], mc[A]);

        if (AB > BC && AB > CA)
        {
            outlier = C; median1 = A; median2 = B;
        }
        else if (CA > AB && CA > BC)
        {
            outlier = B; median1 = A; median2 = C;
        }
        else if (BC > AB && BC > CA)
        {
            outlier = A;  median1 = B; median2 = C;
        }

        top = outlier;                          // The obvious choice

        dist = cv_lineEquation(mc[median1], mc[median2], mc[outlier]);  // Get the Perpendicular distance of the outlier from the longest side          
        slope = cv_lineSlope(mc[median1], mc[median2], align);      // Also calculate the slope of the longest side

                                                                    // Now that we have the orientation of the line formed median1 & median2 and we also have the position of the outlier w.r.t. the line
                                                                    // Determine the 'right' and 'bottom' markers

        if (align == 0)
        {
            bottom = median1;
            right = median2;
        }
        else if (slope < 0 && dist < 0)     // Orientation - North
        {
            bottom = median1;
            right = median2;
            orientation = CV_QR_NORTH;
        }
        else if (slope > 0 && dist < 0)     // Orientation - East
        {
            right = median1;
            bottom = median2;
            orientation = CV_QR_EAST;
        }
        else if (slope < 0 && dist > 0)     // Orientation - South          
        {
            right = median1;
            bottom = median2;
            orientation = CV_QR_SOUTH;
        }

        else if (slope > 0 && dist > 0)     // Orientation - West
        {
            bottom = median1;
            right = median2;
            orientation = CV_QR_WEST;
        }


        // To ensure any unintended values do not sneak up when QR code is not present
        float area_top, area_right, area_bottom;

        if (top < contours.size() && right < contours.size() && bottom < contours.size() && contourArea(contours[top]) > 10 && contourArea(contours[right]) > 10 && contourArea(contours[bottom]) > 10)
        {

            vector<Point2f> L, M, O, tempL, tempM, tempO;
            Point2f N;

            vector<Point2f> src, dst;       // src - Source Points basically the 4 end co-ordinates of the overlay image
                                            // dst - Destination Points to transform overlay image  

            Mat warp_matrix;

            cv_getVertices(contours, top, slope, tempL);
            cv_getVertices(contours, right, slope, tempM);
            cv_getVertices(contours, bottom, slope, tempO);

            cv_updateCornerOr(orientation, tempL, L);           // Re-arrange marker corners w.r.t orientation of the QR code
            cv_updateCornerOr(orientation, tempM, M);           // Re-arrange marker corners w.r.t orientation of the QR code
            cv_updateCornerOr(orientation, tempO, O);           // Re-arrange marker corners w.r.t orientation of the QR code

            int iflag = getIntersectionPoint(M[1], M[2], O[3], O[2], N);


            src.push_back(L[0]);
            src.push_back(M[1]);
            src.push_back(N);
            src.push_back(O[3]);

            dst.push_back(Point2f(0, 0));
            dst.push_back(Point2f(qr.cols, 0));
            dst.push_back(Point2f(qr.cols, qr.rows));
            dst.push_back(Point2f(0, qr.rows));

            if (src.size() == 4 && dst.size() == 4)         // Failsafe for WarpMatrix Calculation to have only 4 Points with src and dst
            {
                warp_matrix = getPerspectiveTransform(src, dst);
                warpPerspective(image, qr_raw, warp_matrix, Size(qr.cols, qr.rows)); 
                copyMakeBorder(qr_raw, qr, 10, 10, 10, 10, BORDER_CONSTANT, Scalar(255, 255, 255));

                cvtColor(qr, qr_gray, CV_RGB2GRAY);
                threshold(qr_gray, qr_thres, 127, 255, CV_THRESH_BINARY); 

                //threshold(qr_gray, qr_thres, 0, 255, CV_THRESH_OTSU);
                //for( int d=0 ; d < 4 ; d++){  src.pop_back(); dst.pop_back(); }


            }

            //Draw contours on the image
            drawContours(image, contours, top, Scalar(255, 200, 0), 2, 8, hierarchy, 0);
            drawContours(image, contours, right, Scalar(0, 0, 255), 2, 8, hierarchy, 0);
            drawContours(image, contours, bottom, Scalar(255, 0, 100), 2, 8, hierarchy, 0);

            // Insert Debug instructions here
            if (DBG == 1)
            {
                // Debug Prints
                // Visualizations for ease of understanding
                if (slope > 5)
                    circle(traces, Point(10, 20), 5, Scalar(0, 0, 255), -1, 8, 0);
                else if (slope < -5)
                    circle(traces, Point(10, 20), 5, Scalar(255, 255, 255), -1, 8, 0);

                // Draw contours on Trace image for analysis    
                drawContours(traces, contours, top, Scalar(255, 0, 100), 1, 8, hierarchy, 0);
                drawContours(traces, contours, right, Scalar(255, 0, 100), 1, 8, hierarchy, 0);
                drawContours(traces, contours, bottom, Scalar(255, 0, 100), 1, 8, hierarchy, 0);

                // Draw points (4 corners) on Trace image for each Identification marker    
                circle(traces, L[0], 2, Scalar(255, 255, 0), -1, 8, 0);
                circle(traces, L[1], 2, Scalar(0, 255, 0), -1, 8, 0);
                circle(traces, L[2], 2, Scalar(0, 0, 255), -1, 8, 0);
                circle(traces, L[3], 2, Scalar(128, 128, 128), -1, 8, 0);

                circle(traces, M[0], 2, Scalar(255, 255, 0), -1, 8, 0);
                circle(traces, M[1], 2, Scalar(0, 255, 0), -1, 8, 0);
                circle(traces, M[2], 2, Scalar(0, 0, 255), -1, 8, 0);
                circle(traces, M[3], 2, Scalar(128, 128, 128), -1, 8, 0);

                circle(traces, O[0], 2, Scalar(255, 255, 0), -1, 8, 0);
                circle(traces, O[1], 2, Scalar(0, 255, 0), -1, 8, 0);
                circle(traces, O[2], 2, Scalar(0, 0, 255), -1, 8, 0);
                circle(traces, O[3], 2, Scalar(128, 128, 128), -1, 8, 0);

                // Draw point of the estimated 4th Corner of (entire) QR Code
                circle(traces, N, 2, Scalar(255, 255, 255), -1, 8, 0);

                // Draw the lines used for estimating the 4th Corner of QR Code
                line(traces, M[1], N, Scalar(0, 0, 255), 1, 8, 0);
                line(traces, O[3], N, Scalar(0, 0, 255), 1, 8, 0);


                // Show the Orientation of the QR Code wrt to 2D Image Space
                int fontFace = FONT_HERSHEY_PLAIN;

                if (orientation == CV_QR_NORTH)
                {
                    putText(traces, "NORTH", Point(20, 30), fontFace, 1, Scalar(0, 255, 0), 1, 8);
                }
                else if (orientation == CV_QR_EAST)
                {
                    putText(traces, "EAST", Point(20, 30), fontFace, 1, Scalar(0, 255, 0), 1, 8);
                }
                else if (orientation == CV_QR_SOUTH)
                {
                    putText(traces, "SOUTH", Point(20, 30), fontFace, 1, Scalar(0, 255, 0), 1, 8);
                }
                else if (orientation == CV_QR_WEST)
                {
                    putText(traces, "WEST", Point(20, 30), fontFace, 1, Scalar(0, 255, 0), 1, 8);
                }

                // Debug Prints
                imwrite("out.bmp", qr_thres);
            }


        }
    }

    imshow("Image", image);
    imshow("Traces", traces);
    imshow("QR code", qr_thres);

    key = waitKey(2);   // OPENCV: wait for 1ms before accessing next frame

}   // End of 'while' loop

return 0;
}

// End of Main Loop
//--------------------------------------------------------------------------------------


// Routines used in Main loops

// Function: Routine to get Distance between two points
// Description: Given 2 points, the function returns the distance

float cv_distance(Point2f P, Point2f Q)
{
    return sqrt(pow(abs(P.x - Q.x), 2) + pow(abs(P.y - Q.y), 2));
}


// Function: Perpendicular Distance of a Point J from line formed by Points L and M; Equation of the line ax+by+c=0
// Description: Given 3 points, the function derives the line quation of the first two points,
//    calculates and returns the perpendicular distance of the the 3rd point from this line.

float cv_lineEquation(Point2f L, Point2f M, Point2f J)
{
    float a, b, c, pdist;

    a = -((M.y - L.y) / (M.x - L.x));
    b = 1.0;
    c = (((M.y - L.y) / (M.x - L.x)) * L.x) - L.y;

    // Now that we have a, b, c from the equation ax + by + c, time to substitute (x,y) by values from the Point J

    pdist = (a * J.x + (b * J.y) + c) / sqrt((a * a) + (b * b));
    return pdist;
}

// Function: Slope of a line by two Points L and M on it; Slope of line, S = (x1 -x2) / (y1- y2)
// Description: Function returns the slope of the line formed by given 2 points, the alignement flag
//    indicates the line is vertical and the slope is infinity.

float cv_lineSlope(Point2f L, Point2f M, int& alignement)
{
    float dx, dy;
    dx = M.x - L.x;
    dy = M.y - L.y;

    if (dy != 0)
    {
        alignement = 1;
        return (dy / dx);
    }
    else                // Make sure we are not dividing by zero; so use 'alignement' flag
    {
        alignement = 0;
        return 0.0;
    }
}



// Function: Routine to calculate 4 Corners of the Marker in Image Space using Region partitioning
// Theory: OpenCV Contours stores all points that describe it and these points lie the perimeter of the polygon.
//  The below function chooses the farthest points of the polygon since they form the vertices of that polygon,
//  exactly the points we are looking for. To choose the farthest point, the polygon is divided/partitioned into
//  4 regions equal regions using bounding box. Distance algorithm is applied between the centre of bounding box
//  every contour point in that region, the farthest point is deemed as the vertex of that region. Calculating
//  for all 4 regions we obtain the 4 corners of the polygon ( - quadrilateral).
void cv_getVertices(vector<vector<Point> > contours, int c_id, float slope, vector<Point2f>& quad)
{
    Rect box;
    box = boundingRect(contours[c_id]);

    Point2f M0, M1, M2, M3;
    Point2f A, B, C, D, W, X, Y, Z;

    A = box.tl();
    B.x = box.br().x;
    B.y = box.tl().y;
    C = box.br();
    D.x = box.tl().x;
    D.y = box.br().y;


    W.x = (A.x + B.x) / 2;
    W.y = A.y;

    X.x = B.x;
    X.y = (B.y + C.y) / 2;

    Y.x = (C.x + D.x) / 2;
    Y.y = C.y;

    Z.x = D.x;
    Z.y = (D.y + A.y) / 2;

    float dmax[4];
    dmax[0] = 0.0;
    dmax[1] = 0.0;
    dmax[2] = 0.0;
    dmax[3] = 0.0;

    float pd1 = 0.0;
    float pd2 = 0.0;

    if (slope > 5 || slope < -5)
    {

        for (int i = 0; i < contours[c_id].size(); i++)
        {
            pd1 = cv_lineEquation(C, A, contours[c_id][i]); // Position of point w.r.t the diagonal AC 
            pd2 = cv_lineEquation(B, D, contours[c_id][i]); // Position of point w.r.t the diagonal BD

            if ((pd1 >= 0.0) && (pd2 > 0.0))
            {
                cv_updateCorner(contours[c_id][i], W, dmax[1], M1);
            }
            else if ((pd1 > 0.0) && (pd2 <= 0.0))
            {
                cv_updateCorner(contours[c_id][i], X, dmax[2], M2);
            }
            else if ((pd1 <= 0.0) && (pd2 < 0.0))
            {
                cv_updateCorner(contours[c_id][i], Y, dmax[3], M3);
            }
            else if ((pd1 < 0.0) && (pd2 >= 0.0))
            {
                cv_updateCorner(contours[c_id][i], Z, dmax[0], M0);
            }
            else
                continue;
        }
    }
    else
    {
        int halfx = (A.x + B.x) / 2;
        int halfy = (A.y + D.y) / 2;

        for (int i = 0; i < contours[c_id].size(); i++)
        {
            if ((contours[c_id][i].x < halfx) && (contours[c_id][i].y <= halfy))
            {
                cv_updateCorner(contours[c_id][i], C, dmax[2], M0);
            }
            else if ((contours[c_id][i].x >= halfx) && (contours[c_id][i].y < halfy))
            {
                cv_updateCorner(contours[c_id][i], D, dmax[3], M1);
            }
            else if ((contours[c_id][i].x > halfx) && (contours[c_id][i].y >= halfy))
            {
                cv_updateCorner(contours[c_id][i], A, dmax[0], M2);
            }
            else if ((contours[c_id][i].x <= halfx) && (contours[c_id][i].y > halfy))
            {
                cv_updateCorner(contours[c_id][i], B, dmax[1], M3);
            }
        }
    }

    quad.push_back(M0);
    quad.push_back(M1);
    quad.push_back(M2);
    quad.push_back(M3);

}

// Function: Compare a point if it more far than previously recorded farthest distance
// Description: Farthest Point detection using reference point and baseline distance
void cv_updateCorner(Point2f P, Point2f ref, float& baseline, Point2f& corner)
{
    float temp_dist;
    temp_dist = cv_distance(P, ref);

    if (temp_dist > baseline)
    {
        baseline = temp_dist;           // The farthest distance is the new baseline
        corner = P;                     // P is now the farthest point
    }

}

// Function: Sequence the Corners wrt to the orientation of the QR Code
void cv_updateCornerOr(int orientation, vector<Point2f> IN, vector<Point2f> &OUT)
{
    Point2f M0, M1, M2, M3;
    if (orientation == CV_QR_NORTH)
    {
        M0 = IN[0];
        M1 = IN[1];
        M2 = IN[2];
        M3 = IN[3];
    }
    else if (orientation == CV_QR_EAST)
    {
        M0 = IN[1];
        M1 = IN[2];
        M2 = IN[3];
        M3 = IN[0];
    }
    else if (orientation == CV_QR_SOUTH)
    {
        M0 = IN[2];
        M1 = IN[3];
        M2 = IN[0];
        M3 = IN[1];
    }
    else if (orientation == CV_QR_WEST)
    {
        M0 = IN[3];
        M1 = IN[0];
        M2 = IN[1];
        M3 = IN[2];
    }

    OUT.push_back(M0);
    OUT.push_back(M1);
    OUT.push_back(M2);
    OUT.push_back(M3);
}

// Function: Get the Intersection Point of the lines formed by sets of two points
bool getIntersectionPoint(Point2f a1, Point2f a2, Point2f b1, Point2f b2, Point2f& intersection)
{
    Point2f p = a1;
    Point2f q = b1;
    Point2f r(a2 - a1);
    Point2f s(b2 - b1);

    if (cross(r, s) == 0) { return false; }

    float t = cross(q - p, s) / cross(r, s);

    intersection = p + t * r;
    return true;
}

float cross(Point2f v1, Point2f v2)
{
    return v1.x*v2.y - v1.y*v2.x;
}

these are the QR codes that the program should recognize from the video:

https://drive.google.com/file/d/1uqZSJCpBML8vb7AuDU9KpTeXHt-3IAPi/view?usp=sharing

https://drive.google.com/open?id=1VDj2DalqB-0lWaFwVC_Nz6nT-8rUZ1NA

https://drive.google.com/open?id=12L8V2WuHJ-xomOgQqn1CIUh2s8BD24OZ

https://drive.google.com/open?id=1CRCBrczOMzdSLuwb-mYoU5d_iB2MIVxJ