Inner angle between two lines

问题:

I have two lines: Line1 and Line2. Each line is defined by two points (P1L1(x1, y1), P2L1(x2, y2) and P1L1(x1, y1), P2L3(x2, y3)). I want to know the inner angle defined by these two lines.

For do it I calculate the angle of each line with the abscissa:

double theta1 = atan(m1) * (180.0 / PI);
double theta2 = atan(m2) * (180.0 / PI);

After to know the angle I calculate the following:

double angle = abs(theta2 - theta1);

The problem or doubt that I have is: sometimes I get the correct angle but sometimes I get the complementary angle (for me outer). How can I know when subtract 180º to know the inner angle? There is any algorithm better to do that? Because I tried some methods: dot product, following formula:

result = (m1 - m2) / (1.0 + (m1 * m2));

But always I have the same problem; I never known when I have the outer angle or the inner angle!

回答1:

I think what you're looking for is the inner product (you may also want to look over the dot product entry) of the two angles. In your case, that's given by:

float dx21 = x2-x1;
float dx31 = x3-x1;
float dy21 = y2-y1;
float dy31 = y3-y1;
float m12 = sqrt( dx21*dx21 + dy21*dy21 );
float m13 = sqrt( dx31*dx31 + dy31*dy31 );
float theta = acos( (dx21*dx31 + dy21*dy31) / (m12 * m13) );

Answer is in radians.

EDIT: Here's a complete implementation. Substitute the problematic values in p1, p2, and p3 and let me know what you get. The point p1 is the vertex where the two lines intersect, in accordance with your definition of the two lines.

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

template <typename T> class Vector2D
{
private:
    T x;
    T y;

public:
    explicit Vector2D(const T& x=0, const T& y=0) : x(x), y(y) {}
    Vector2D(const Vector2D<T>& src) : x(src.x), y(src.y) {}
    virtual ~Vector2D() {}

    // Accessors
    inline T X() const { return x; }
    inline T Y() const { return y; }
    inline T X(const T& x) { this->x = x; }
    inline T Y(const T& y) { this->y = y; }

    // Vector arithmetic
    inline Vector2D<T> operator-() const
        { return Vector2D<T>(-x, -y); }

    inline Vector2D<T> operator+() const
        { return Vector2D<T>(+x, +y); }

    inline Vector2D<T> operator+(const Vector2D<T>& v) const
        { return Vector2D<T>(x+v.x, y+v.y); }

    inline Vector2D<T> operator-(const Vector2D<T>& v) const
        { return Vector2D<T>(x-v.x, y-v.y); }

    inline Vector2D<T> operator*(const T& s) const
        { return Vector2D<T>(x*s, y*s); }

    // Dot product
    inline T operator*(const Vector2D<T>& v) const
        { return x*v.x + y*v.y; }

    // l-2 norm
    inline T norm() const { return sqrt(x*x + y*y); }

    // inner angle (radians)
    static T angle(const Vector2D<T>& v1, const Vector2D<T>& v2)
    {
        return acos( (v1 * v2) / (v1.norm() * v2.norm()) );
    }
};

int main()
{
    Vector2D<double> p1(215, 294);
    Vector2D<double> p2(174, 228);
    Vector2D<double> p3(303, 294);

    double rad = Vector2D<double>::angle(p2-p1, p3-p1);
    double deg = rad * 180.0 / M_PI;

    std::cout << "rad = " << rad << "\tdeg = " << deg << std::endl;

    p1 = Vector2D<double>(153, 457);
    p2 = Vector2D<double>(19, 457);
    p3 = Vector2D<double>(15, 470);

    rad = Vector2D<double>::angle(p2-p1, p3-p1);
    deg = rad * 180.0 / M_PI;

    std::cout << "rad = " << rad << "\tdeg = " << deg << std::endl;

    return 0;
}

The code above yields:

rad = 2.12667   deg = 121.849
rad = 0.0939257 deg = 5.38155

回答2:

if (result > 180)
{
     result = 360 - result;
}

That way it will always be the inner angle. Just add it after you get result.

回答3:

If you want in between angle in 0 degree to 360 degree then use following code; Its fully tested and functional:

static inline CGFloat angleBetweenLinesInRadians(CGPoint line1Start, CGPoint line1End, CGPoint line2Start, CGPoint line2End) {
double angle1 = atan2(line1Start.y-line1End.y, line1Start.x-line1End.x);
double angle2 = atan2(line2Start.y-line2End.y, line2Start.x-line2End.x);
double result = (angle2-angle1) * 180 / 3.14;
if (result<0) {
    result+=360;
}
return result;

}

Note: Rotation will be clockwise;

回答4:

Inner angle between 2 vectors (v1, v2) = arc cos ( inner product(v1,v2) / (module(v1) * module(v2)) ).

Where inner product(v1,v2) = xv1*xv2 + yv1*yv2

module(v) = sqrt(pow(xv,2) + pow(yv,2))

So, the answer of your question is implemented on the following example:

#define PI   3.14159258

int main()
{
    double x1,y1,x2,y2,y3;
    double m1, m2;
    double mod1, mod2, innerp, angle;

    cout << "x1 :";
    cin >> x1;
    cout << "y1 :";
    cin >> y1;
    cout << "x2 :";
    cin >> x2;
    cout << "y2 :";
    cin >> y2;
    cout << "y3 :";
    cin >> y3;

    m1 = atan((y2-y1)/(x2-x1)) * 180 / PI;
    m2 = atan((y3-y1)/(x2-x1)) * 180 / PI;

    mod1   = sqrt(pow(y2-y1,2)+pow(x2-x1,2));
    mod2   = sqrt(pow(y3-y1,2)+pow(x2-x1,2));
    innerp = (x2-x1)*(x2-x1) + (y2-y1)*(y3-y1);
    angle  = acos(innerp / (mod1 * mod2)) * 180 / PI;

    cout << "m1 : " << m1 << endl;
    cout << "m2 : " << m2 << endl;
    cout << "angle : " << angle << endl;
}

回答5:

The whole point is much easier than the given answers:

When you use atan(slope) you lose (literally) one bit of information, that is there are exactly two angles (theta) and (theta+PI) in the range (0..2*PI), which give the same value for the function tan().

Just use atan2(deltax, deltay) and you get the right angle. For instance

atan2(1,1) == PI/4
atan2(-1,-1) == 5*PI/4

Then subtract, take absolute value, and if greater than PI subtract from 2*PI.

回答6:

If you use abolute value you will always get the acute angle. That is tangent theta = abs value of m1-m2 over (1 +m1 * m2). If you take inverse tangent your answer will be in radians or degrees however the calculator is set. Sorry this isnt programming lingo, I am a math teacher, not a programmer...

回答7:

Getting the outer angle vs the inner angle is determined entirely by the order of your subtractions (think about it). You need to subtract the smaller theta from the larger in order to reliably always get the inner angle. You also probably want to use the atan2 function because of the type of data you're expecting.

回答8:

I hope I understand your question correctly as wanting the acute angle rather than the obtuse angle of the intersection of two lines. Am I correct?

Acute and obtuse angles of an intersection are 180 deg complements of each other. i.e.

 acute + obtuse = PI.

http://www.mathworks.com/access/helpdesk/help/techdoc/ref/atan.html exhibits that an atan is asymptotic at +/- pi/2.

Therefore, the max difference between two results of atan is pi or 180 deg, whether you use the +/- notation or positive 0 to pi notation of a gradient.

Consider the following pseudocode:

acuteAngle(m1, m2){
  a = atan(m1) - atan(m2);

  // if obtuse get the complementary acute angle:
  if (a>PI/2) 
    a = PI - a;
  return a;
} 

The function acuteAngle illustrates what you need to do, mathematically.

However, it cannot be used for values of angles in the neighbourhood of PI/2 because binary comparisons of angles with results in that neighbourhood is questionable whether an obtuse or acute angle is represented.

Therefore, we have to compare the coordinates of the points of the two lines. We find out whether the 3rd line formed from [(x2,y2)(x3,y3)] is shorter, equal or longer than the hypothetical hypotenuse.

By virtue of Pythagoras' theorem, A hypotenuse is formed if the angle is exactly PI/2 or 90 deg. Let's call his hypothetical hypotenuse line L3Hypo.

By geometrical visualisation in your mind,

• If the 3rd line is longer than L3Hypo, the angle is obtuse.
• If shorter, the angle is acute.
• Otherwise, perfect 90.

Therefore,

L1.lengthSquared = sq(x2-x1) + sq(y2-y1)
L2.lengthSquared = sq(x3-x1) + sq(y3-y1)
L3Hypo.lengthSquared = L1.lengthSquared + L2.lengthSquared
L3.lengthSquared = sq(x3-x2) + sq(y3-y2)

Therefore, the following pseudo-code,

struct Point{
  double x, y;
}

// no need to struct, for clarity only
struct Line{
  double lengthSquared;
}

#define sq(n) (n*n)
int isObtuse(Point P1, P2, P3){
  Line L1, L2, L3, L3Hypo;

  L1.lengthSquared = sq(P2.x-P1.x) + sq(P2.y-P1.y);
  L2.lengthSquared = sq(P3.x-P1.x) + sq(P3.y-P1.y);
  L3Hypo.lengthSquared = L1.lengthSquared + L2.lengthSquared;
  L3.lengthSquared = sq(P3.x-P2.x) + sq(P3.y-P2.y);

  if (L3>L3Hypo) return 1; //obtuse
  else if (L3<L3Hypo) return -1; //acute
  else return 0;
}

Presuming you already have the function getGradient(Point P, Q):

double m1m2 = getGradient(P1,P2);
double m1m3 = getGradient(P1,P3);
double a = Abs(atan(m1m2) - atan(m1m3));
if (isObtuse(P1, P2, P3)>0)
  a = PI - a;

I may have committed some typo mistakes in the pseudo-code (hopefully not) but I demonstrated the gist of the concept. If so, someone could be so kind to edit away the typos.

Further However, after mulling over it, I find that the struggle for precision pivots on its weakest link due to the directive

#define PI 3.14159blah..blah..blah.

So, we might as well save all the trouble and simply do this:

double m1m2 = getGradient(P1,P2);
double m1m3 = getGradient(P1,P3);
double a = Abs(atan(m1m2) - atan(m1m3));
double b = PI - a;
return min(a, b);//the smaller of the two is the acute