Edges on polygon outlines not always correct

Question

I'm using the algorithm below to generate quads which are then rendered to make an outline like this

http://img810.imageshack.us/img810/8530/uhohz.png

The problem as seen on the image is that sometimes the lines are too thin when they should always be the same width. My algorithm finds the 4 vertices for the first one then the top 2 vertices of the next ones are the bottom 2 of the previous. This creates connected lines, but it seems to not always work. How could I fix this?

This is my algorithm:

 void OGLENGINEFUNCTIONS::GenerateLinePoly(const std::vector<std::vector<GLdouble>> &input,
                          std::vector<GLfloat> &output, int width)
 {
     output.clear();

     if(input.size() < 2)
     {
         return;
     }

     int temp;
     float dirlen;
     float perplen;
     POINTFLOAT start;
     POINTFLOAT end;
     POINTFLOAT dir;
     POINTFLOAT ndir;
     POINTFLOAT perp;
     POINTFLOAT nperp;

     POINTFLOAT perpoffset;
     POINTFLOAT diroffset;

     POINTFLOAT p0, p1, p2, p3;

     for(unsigned int i = 0; i < input.size() - 1; ++i)
     {

         start.x = static_cast<float>(input[i][0]);
         start.y = static_cast<float>(input[i][1]);

         end.x = static_cast<float>(input[i + 1][0]);
         end.y = static_cast<float>(input[i + 1][1]);

         dir.x = end.x - start.x;
         dir.y = end.y - start.y;

         dirlen = sqrt((dir.x * dir.x) + (dir.y * dir.y));

         ndir.x = static_cast<float>(dir.x * 1.0 / dirlen);
         ndir.y = static_cast<float>(dir.y * 1.0 / dirlen);

         perp.x = dir.y;
         perp.y = -dir.x;

         perplen = sqrt((perp.x * perp.x) + (perp.y * perp.y));

         nperp.x = static_cast<float>(perp.x * 1.0 / perplen);
         nperp.y = static_cast<float>(perp.y * 1.0 / perplen);

         perpoffset.x = static_cast<float>(nperp.x * width * 0.5);
         perpoffset.y = static_cast<float>(nperp.y * width * 0.5);

         diroffset.x = static_cast<float>(ndir.x * 0 * 0.5);
         diroffset.y = static_cast<float>(ndir.y * 0 * 0.5);

            // p0 = start + perpoffset - diroffset
            // p1 = start - perpoffset - diroffset
            // p2 = end + perpoffset + diroffset
            // p3 = end - perpoffset + diroffset 

         p0.x = start.x + perpoffset.x - diroffset.x;
         p0.y = start.y + perpoffset.y - diroffset.y;

         p1.x = start.x - perpoffset.x - diroffset.x;
         p1.y = start.y - perpoffset.y - diroffset.y;

         if(i > 0)
         {
             temp = (8 * (i - 1));
             p2.x = output[temp + 2];
             p2.y = output[temp + 3];
             p3.x = output[temp + 4];
             p3.y = output[temp + 5];

         }
         else
         {
             p2.x = end.x + perpoffset.x + diroffset.x;
             p2.y = end.y + perpoffset.y + diroffset.y;

             p3.x = end.x - perpoffset.x + diroffset.x;
             p3.y = end.y - perpoffset.y + diroffset.y;
         }



         output.push_back(p2.x);
         output.push_back(p2.y);
         output.push_back(p0.x);
         output.push_back(p0.y);
         output.push_back(p1.x);
         output.push_back(p1.y);
         output.push_back(p3.x);
         output.push_back(p3.y);

     }
 }

---

Edit:

 POINTFLOAT multiply(const POINTFLOAT &a, float b)
 {
     POINTFLOAT result;
     result.x = a.x * b;
     result.y = a.y * b;
     return result;
 }

 POINTFLOAT normalize(const POINTFLOAT &a)
 {
     return multiply(a, 1.0f / sqrt(a.x * a.x + a.y * a.y));
 }


 POINTFLOAT slerp2d( const POINTFLOAT v0, 
                     const POINTFLOAT v1, float t )
 {
     float dot = (v0.x * v1.x + v1.y * v1.y);
     if( dot < -1.0f ) dot = -1.0f;
     if( dot > 1.0f ) dot = 1.0f;

     float theta_0 = acos( dot );
     float theta = theta_0 * t;

     POINTFLOAT v2;
     v2.x = -v0.y;
     v2.y = v0.x;

     POINTFLOAT result;
     result.x = v0.x * cos(theta) + v2.x * sin(theta);
     result.y = v0.y * cos(theta) + v2.y * sin(theta);

     return result;
 }

 void OGLENGINEFUNCTIONS::GenerateLinePoly(const std::vector<std::vector<GLdouble> > &input,
                          std::vector<GLfloat> &output, int width)
 {
     output.clear();

     if(input.size() < 2)
     {
         return;
     }

     float w = width / 2.0f;

     //glBegin(GL_TRIANGLES);
     for( size_t i = 0; i < input.size()-1; ++i )
     {
         POINTFLOAT cur;
         cur.x = input[i][0];
         cur.y = input[i][1];


         POINTFLOAT nxt;
         nxt.x = input[i+1][0];
         nxt.y = input[i+1][1];

         POINTFLOAT b;
         b.x = nxt.x - cur.x;
         b.y = nxt.y - cur.y;

         b = normalize(b);



         POINTFLOAT b_perp;
         b_perp.x = -b.y;
         b_perp.y = b.x;


         POINTFLOAT p0;
         POINTFLOAT p1;
         POINTFLOAT p2;
         POINTFLOAT p3;

         p0.x = cur.x + b_perp.x * w;
         p0.y = cur.y + b_perp.y * w;

         p1.x = cur.x - b_perp.x * w;
         p1.y = cur.y - b_perp.y * w;

         p2.x = nxt.x + b_perp.x * w;
         p2.y = nxt.y + b_perp.y * w;

         p3.x = nxt.x - b_perp.x * w;
         p3.y = nxt.y - b_perp.y * w;

         output.push_back(p0.x);
         output.push_back(p0.y);
         output.push_back(p1.x);
         output.push_back(p1.y);
         output.push_back(p2.x);
         output.push_back(p2.y);

         output.push_back(p2.x);
         output.push_back(p2.y);
         output.push_back(p1.x);
         output.push_back(p1.y);
         output.push_back(p3.x);
         output.push_back(p3.y);



         // only do joins when we have a prv
         if( i == 0 ) continue;

         POINTFLOAT prv;
         prv.x = input[i-1][0];
         prv.y = input[i-1][1];

         POINTFLOAT a;
         a.x = prv.x - cur.x;
         a.y = prv.y - cur.y;

         a = normalize(a);

         POINTFLOAT a_perp;
         a_perp.x = a.y;
         a_perp.y = -a.x;

         float det = a.x * b.y - b.x * a.y;
         if( det > 0 )
         {
             a_perp.x = -a_perp.x;
             a_perp.y = -a_perp.y;

             b_perp.x = -b_perp.x;
             b_perp.y = -b_perp.y;
         }

         // TODO: do inner miter calculation

         // flip around normals and calculate round join points
         a_perp.x = -a_perp.x;
         a_perp.y = -a_perp.y;

         b_perp.x = -b_perp.x;
         b_perp.y = -b_perp.y;

         size_t num_pts = 4;

         std::vector< POINTFLOAT> round( 1 + num_pts + 1 );
         POINTFLOAT nc;
         nc.x = cur.x + (a_perp.x * w);
         nc.y = cur.y + (a_perp.y * w);

         round.front() = nc;

         nc.x = cur.x + (b_perp.x * w);
         nc.y = cur.y + (b_perp.y * w);

         round.back() = nc;

         for( size_t j = 1; j < num_pts+1; ++j )
         {
             float t = (float)j / (float)(num_pts + 1);
             if( det > 0 )
             {
                 POINTFLOAT nin;
                 nin = slerp2d( b_perp, a_perp, 1.0f-t );
                 nin.x *= w;
                 nin.y *= w;

                 nin.x += cur.x;
                 nin.y += cur.y;

                 round[j] = nin;
             }
             else
             {
                 POINTFLOAT nin;
                 nin = slerp2d( a_perp, b_perp, t );
                 nin.x *= w;
                 nin.y *= w;

                 nin.x += cur.x;
                 nin.y += cur.y;

                 round[j] = nin;
             }
         }

         for( size_t j = 0; j < round.size()-1; ++j )
         {

             output.push_back(cur.x);
             output.push_back(cur.y);

             if( det > 0 )
             {
                 output.push_back(round[j + 1].x);
                 output.push_back(round[j + 1].y);
                 output.push_back(round[j].x);
                 output.push_back(round[j].y);
             }
             else
             {

                 output.push_back(round[j].x);
                 output.push_back(round[j].y);

                 output.push_back(round[j + 1].x);
                 output.push_back(round[j + 1].y);
             }
         }
     }
 }

Answer 1

Requires Eigen as written, but the core operations should map easily to whatever vector class you're using.

// v0 and v1 are normalized
// t can vary between 0 and 1
// http://number-none.com/product/Understanding%20Slerp,%20Then%20Not%20Using%20It/
Vector2f slerp2d( const Vector2f& v0, const Vector2f& v1, float t )
{
    float dot = v0.dot(v1);
    if( dot < -1.0f ) dot = -1.0f;
    if( dot > 1.0f ) dot = 1.0f;

    float theta_0 = acos( dot );
    float theta = theta_0 * t;

    Vector2f v2( -v0.y(), v0.x() );

    return ( v0*cos(theta) + v2*sin(theta) );
}


void glPolyline( const vector<Vector2f>& polyline, float width )
{
    if( polyline.size() < 2 ) return;
    float w = width / 2.0f;

    glBegin(GL_TRIANGLES);
    for( size_t i = 0; i < polyline.size()-1; ++i )
    {
        const Vector2f& cur = polyline[ i ];
        const Vector2f& nxt = polyline[i+1];

        Vector2f b = (nxt - cur).normalized();
        Vector2f b_perp( -b.y(), b.x() );

        Vector2f p0( cur + b_perp*w );
        Vector2f p1( cur - b_perp*w );
        Vector2f p2( nxt + b_perp*w );
        Vector2f p3( nxt - b_perp*w );

        // first triangle
        glVertex2fv( p0.data() );
        glVertex2fv( p1.data() );
        glVertex2fv( p2.data() );
        // second triangle
        glVertex2fv( p2.data() );
        glVertex2fv( p1.data() );
        glVertex2fv( p3.data() );

        // only do joins when we have a prv
        if( i == 0 ) continue;

        const Vector2f& prv = polyline[i-1];
        Vector2f a = (prv - cur).normalized();
        Vector2f a_perp( a.y(), -a.x() );

        float det = a.x()*b.y() - b.x()*a.y();
        if( det > 0 )
        {
            a_perp = -a_perp;
            b_perp = -b_perp;
        }

        // TODO: do inner miter calculation

        // flip around normals and calculate round join points
        a_perp = -a_perp;
        b_perp = -b_perp;

        size_t num_pts = 4;
        vector< Vector2f > round( 1 + num_pts + 1 );
        for( size_t j = 0; j <= num_pts+1; ++j )
        {
            float t = (float)j/(float)(num_pts+1);
            if( det > 0 )
                round[j] = cur + (slerp2d( b_perp, a_perp, 1.0f-t ) * w);
            else
                round[j] = cur + (slerp2d( a_perp, b_perp, t ) * w);
        }

        for( size_t j = 0; j < round.size()-1; ++j )
        {
            glVertex2fv( cur.data() );
            if( det > 0 )
            {
                glVertex2fv( round[j+1].data() );
                glVertex2fv( round[j+0].data() );
            }
            else
            {
                glVertex2fv( round[j+0].data() );
                glVertex2fv( round[j+1].data() );
            }
        }
    }
    glEnd();
}

EDIT: Screenshots:

Answer 2

What about:

1. Draw each line up to the inside of the corner
2. Draw an extra line at each corner perpendicular to the angle of the corner

Like this:

alt text http://www.geekops.co.uk/photos/0000-00-02%20%28Forum%20images%29/CorrectAngleDrawing.png

Blue / red represent the two lines you're trying to connect. Dotted green is the extra one you add to smooth the corner. The image above shows the content would be clipped very slightly for sharp corners. If that's a problem, you could extend the two connecting lines further outwards and draw the extra line further out.

[Edit] I've spotted a flaw in my suggestion. You have some concave sections in there which won't work at all well. For those occasions you'll want to do something like drawing a chamfered edge instead:

alt text http://www.geekops.co.uk/photos/0000-00-02%20%28Forum%20images%29/CorrectAngleDrawing2.png

[Edit2] I've done a little debugging on the code I posted previously. The following should be of more use:

    // PolygonOutlineGen.cpp : A small program to calculate 4-point polygons 
// to surround an input polygon.

#include <vector>
#include <math.h>
#include <iostream>
#include <iomanip>

using namespace std;

// Describe some structures etc. so the code will compile without 
// requiring the GL libraries.
typedef double GLdouble;
typedef float GLfloat;
typedef struct POINTFLOAT
{
    float x;
    float y;
} POINTFLOAT;

// A function to generate two coordinates representing the start and end
// of a line perpendicular to start/end, offset by 'width' units.
void GenerateOffsetLineCoords(
    POINTFLOAT start, 
    POINTFLOAT end, 
    int width,
    POINTFLOAT& perpStart,
    POINTFLOAT& perpEnd)
{
    float dirlen;
    POINTFLOAT dir;
    POINTFLOAT ndir;
    POINTFLOAT nperp;
    POINTFLOAT perpoffset;

    // Work out the offset for a parallel line which is space outwards by 'width' units
    dir.x = end.x - start.x;
    dir.y = end.y - start.y;
    dirlen = sqrt((dir.x * dir.x) + (dir.y * dir.y));
    ndir.x = static_cast<float>(dir.x * 1.0 / dirlen);
    ndir.y = static_cast<float>(dir.y * 1.0 / dirlen);
    nperp.x = -ndir.y;
    nperp.y = ndir.x;
    perpoffset.x = static_cast<float>(nperp.x * width);
    perpoffset.y = static_cast<float>(nperp.y * width);

    // Calculate the offset coordinates for the new line
    perpStart.x = start.x + perpoffset.x;
    perpStart.y = start.y + perpoffset.y;
    perpEnd.x = end.x + perpoffset.x;
    perpEnd.y = end.y + perpoffset.y;
}

// Function to generate quads of coordinate pairs to surround the 'input'
// polygon.
void GenerateLinePoly(const std::vector<std::vector<GLdouble>> &input,
    std::vector<GLfloat> &output, int width)
{
    // Make sure we have something to produce an outline for and that it's not contaminated with previous results
    output.clear();
    if(input.size() < 2)
    {
        return;
    }

    // Storage for the pairs of lines which form sections of the outline
    POINTFLOAT line1_start;
    POINTFLOAT line1_end;
    POINTFLOAT line2_start;
    POINTFLOAT line2_end;

    // Storage for the outer edges of the quads we'll be generating
    POINTFLOAT line1offset_start;
    POINTFLOAT line1offset_end;
    POINTFLOAT line2offset_start;
    POINTFLOAT line2offset_end;

    // Storage for the line we'll use to make smooth joints between polygon sections.
    POINTFLOAT joininglineoffset_start;
    POINTFLOAT joininglineoffset_end;

    for(unsigned int i = 0; i < input.size() - 2; ++i)
    {
        // Grab the raw line input for the first line or if we've already done one, just re-use the last results
        if( i == 0 )
        {
            line1_start.x = static_cast<float>(input[i][0]);
            line1_start.y = static_cast<float>(input[i][1]);
            line1_end.x = static_cast<float>(input[i + 1][0]);
            line1_end.y = static_cast<float>(input[i + 1][1]);

            GenerateOffsetLineCoords(line1_start, line1_end, width, line1offset_start, line1offset_end);
        }
        else
        {
            line1_start = line2_start;
            line1offset_start = line2offset_start;
            line1_end = line2_end;
            line1offset_end = line2offset_end;
        }

        // Grab the second line and work out the coords of it's offset 
        line2_start.x = static_cast<float>(input[i+1][0]);
        line2_start.y = static_cast<float>(input[i+1][1]);
        line2_end.x = static_cast<float>(input[i+2][0]);
        line2_end.y = static_cast<float>(input[i+2][1]);
        GenerateOffsetLineCoords(line2_start, line2_end, width, line2offset_start, line2offset_end);

        // Grab the offset for the line which joins the open end
        GenerateOffsetLineCoords(line2offset_start, line1offset_end, width, joininglineoffset_start, joininglineoffset_end);

        // Push line 1 onto the output
        output.push_back(line1_start.x);
        output.push_back(line1_start.y);
        output.push_back(line1_end.x);
        output.push_back(line1_end.y);
        output.push_back(line1offset_end.x);
        output.push_back(line1offset_end.y);
        output.push_back(line1offset_start.x);
        output.push_back(line1offset_start.y);

        // Push the new section onto the output
        output.push_back(line1offset_end.x);
        output.push_back(line1offset_end.y);
        output.push_back(line2offset_start.x);
        output.push_back(line2offset_start.y);
        output.push_back(joininglineoffset_start.x);
        output.push_back(joininglineoffset_start.y);
        output.push_back(joininglineoffset_end.x);
        output.push_back(joininglineoffset_end.y);
    }

    // TODO: Push the remaining line 2 on.

    // TODO: Add one last joining piece between the end and the beginning.
}

int main(int argc, char* argv[])
{
    // Describe some input data
    std::vector<std::vector<GLdouble>> input;
    std::vector<GLdouble> val1; val1.push_back(010.0); val1.push_back(010.0); input.push_back(val1);
    std::vector<GLdouble> val2; val2.push_back(050.0); val2.push_back(100.0); input.push_back(val2);
    std::vector<GLdouble> val3; val3.push_back(100.0); val3.push_back(100.0); input.push_back(val3);
    std::vector<GLdouble> val4; val4.push_back(010.0); val4.push_back(010.0); input.push_back(val4);

    // Generate the quads required to outline the shape
    std::vector<GLfloat> output;
    GenerateLinePoly(input, output, 5);

    // Dump the output as pairs of coordinates, grouped into the quads they describe
    cout << setiosflags(ios::fixed) << setprecision(1);
    for(unsigned int i=0; i < output.size(); i++)
    {
       if( (i > 0) && ((i)%2==0) ) { cout << endl; }
       if( (i > 0) && ((i)%8==0) ) { cout << endl; }
       cout << setw(7) << output[i];
    }
    cout << endl;
    return 0;
}

..which looks like it works for convex polygons as far as I can see :-)

Answer 3

This code renders correct SVG instead of incorrect one . It is simpler than genpfault's solution, with the advantage of having less quads to render.

Every connection here is rendered as in the last picture in Jon Cage 's answer.

Answer 4

Ah, I see now. It's because you are reusing your old vertices, which aren't necessarily parallel to your new ones.

Just go through your code with a simple example by hand, where your input points take a sharp 90 degree turn. The old vertices will be parallel to dir while the new ones will be perpendicular. If you have points on the line that are close enough together then you'll get strange behaviour like you see in your picture.

There's no "simple" solution to getting uniform width lines, but things would look better if you just render the lines a pair at a time (ie get rid of the i > 0 case). That will give you some ugly sharp corners, but you won't get any thin lines.

Answer 5

You're reversing the orientation between the first segment and the rest. The block where you pull previous values out of the output vector should set the p0 and p1 points and you should calculate p2 and p3 based on the endpoints every time.

ie it should be:

     if(i == 0)
     {
         p0.x = start.x + perpoffset.x - diroffset.x;
         p0.y = start.y + perpoffset.y - diroffset.y;

         p1.x = start.x - perpoffset.x - diroffset.x;
         p1.y = start.y - perpoffset.y - diroffset.y;
     }
     else
     {
         temp = (8 * (i - 1));
         p0.x = output[temp + 0];
         p0.y = output[temp + 1];
         p1.x = output[temp + 6];
         p1.y = output[temp + 7];

     }

     p2.x = end.x + perpoffset.x + diroffset.x;
     p2.y = end.y + perpoffset.y + diroffset.y;

     p3.x = end.x - perpoffset.x + diroffset.x;
     p3.y = end.y - perpoffset.y + diroffset.y;

Answer 6

You can not use the offset vectors from the previous segment on the current segment - they are perpendicular to something that has nothing to do with the current segment. Best is to use the same offsets like this:

p0.x = start.x + perpoffset.x;
         p0.y = start.y + perpoffset.y;

         p1.x = start.x - perpoffset.x;
         p1.y = start.y - perpoffset.y;

         p2.x = end.x + perpoffset.x;
         p2.y = end.y + perpoffset.y;

         p3.x = end.x - perpoffset.x;
         p3.y = end.y - perpoffset.y;

And then put a circle at each vertex to round the corners. If round is not the way you want to go, you'll have to change the amount of ndir you add to the offsets - this is dependent on BOTH segments connecting at a vertex, not just one. You need to identify the intersection of the incomming and outgoing offset lines. Start with the above, and do some zoomed in shots with angles like 90 or 120 degrees to get a feel for this. Sorry, no formula handy right now.

Lastly, you do not need to normalize the perp vector. The way you calculate it will produce a unit vector anyway.