我要做的是预先计算三个面法线，

在三维中通过边向量和面法向量的叉乘得到。 通过简单地交换分量和负一个，

对于任意一条边的内/外都是边法线和点到点向量的点积，改变符号。重复其他两(或更多)面。

好处:

在同一个三角形上进行多点测试，很多都是预先计算好的。 早期拒签的常见情况是外分多内分。(如果点分布偏向一侧，可以先测试这一侧。)

bool isInside( float x, float y, float x1, float y1, float x2, float y2, float x3, float y3 ) {
  float l1 = (x-x1)*(y3-y1) - (x3-x1)*(y-y1), 
    l2 = (x-x2)*(y1-y2) - (x1-x2)*(y-y2), 
    l3 = (x-x3)*(y2-y3) - (x2-x3)*(y-y3);
  return (l1>0 && l2>0  && l3>0) || (l1<0 && l2<0 && l3<0);
}

没有比这更有效率的了!三角形的每边都可以有独立的位置和方向，因此需要进行l1、l2和l3三个计算，每个计算需要进行2次乘法。一旦l1, l2和l3是已知的，结果只是一些基本的比较和布尔运算。

我要做的是预先计算三个面法线，

在三维中通过边向量和面法向量的叉乘得到。 通过简单地交换分量和负一个，

对于任意一条边的内/外都是边法线和点到点向量的点积，改变符号。重复其他两(或更多)面。

好处:

在同一个三角形上进行多点测试，很多都是预先计算好的。 早期拒签的常见情况是外分多内分。(如果点分布偏向一侧，可以先测试这一侧。)

下面是一个python解决方案，它是高效的，文档化的，包含三个单元测试。它具有专业级的质量，并且可以以模块的形式放入您的项目中。

import unittest

###############################################################################
def point_in_triangle(point, triangle):
    """Returns True if the point is inside the triangle
    and returns False if it falls outside.
    - The argument *point* is a tuple with two elements
    containing the X,Y coordinates respectively.
    - The argument *triangle* is a tuple with three elements each
    element consisting of a tuple of X,Y coordinates.

    It works like this:
    Walk clockwise or counterclockwise around the triangle
    and project the point onto the segment we are crossing
    by using the dot product.
    Finally, check that the vector created is on the same side
    for each of the triangle's segments.
    """
    # Unpack arguments
    x, y = point
    ax, ay = triangle[0]
    bx, by = triangle[1]
    cx, cy = triangle[2]
    # Segment A to B
    side_1 = (x - bx) * (ay - by) - (ax - bx) * (y - by)
    # Segment B to C
    side_2 = (x - cx) * (by - cy) - (bx - cx) * (y - cy)
    # Segment C to A
    side_3 = (x - ax) * (cy - ay) - (cx - ax) * (y - ay)
    # All the signs must be positive or all negative
    return (side_1 < 0.0) == (side_2 < 0.0) == (side_3 < 0.0)

###############################################################################
class TestPointInTriangle(unittest.TestCase):

    triangle = ((22 , 8),
                (12 , 55),
                (7 , 19))

    def test_inside(self):
        point = (15, 20)
        self.assertTrue(point_in_triangle(point, self.triangle))

    def test_outside(self):
        point = (1, 7)
        self.assertFalse(point_in_triangle(point, self.triangle))

    def test_border_case(self):
        """If the point is exactly on one of the triangle's edges,
        we consider it is inside."""
        point = (7, 19)
        self.assertTrue(point_in_triangle(point, self.triangle))

###############################################################################
if __name__ == "__main__":
    suite = unittest.defaultTestLoader.loadTestsFromTestCase(TestPointInTriangle)
    unittest.TextTestRunner().run(suite)

上面的算法有一个额外的可选图形测试，以确认其有效性:

import random
from matplotlib import pyplot
from triangle_test import point_in_triangle

###############################################################################
# The area #
size_x = 64
size_y = 64

# The triangle #
triangle = ((22 , 8),
            (12 , 55),
            (7 , 19))

# Number of random points #
count_points = 10000

# Prepare the figure #
figure = pyplot.figure()
axes = figure.add_subplot(111, aspect='equal')
axes.set_title("Test the 'point_in_triangle' function")
axes.set_xlim(0, size_x)
axes.set_ylim(0, size_y)

# Plot the triangle #
from matplotlib.patches import Polygon
axes.add_patch(Polygon(triangle, linewidth=1, edgecolor='k', facecolor='none'))

# Plot the points #
for i in range(count_points):
    x = random.uniform(0, size_x)
    y = random.uniform(0, size_y)
    if point_in_triangle((x,y), triangle): pyplot.plot(x, y, '.g')
    else:                                  pyplot.plot(x, y, '.b')

# Save it #
figure.savefig("point_in_triangle.pdf")

制作以下图表:

python中的其他函数，比Developer的方法更快(至少对我来说)，并受到Cédric Dufour解决方案的启发:

def ptInTriang(p_test, p0, p1, p2):       
     dX = p_test[0] - p0[0]
     dY = p_test[1] - p0[1]
     dX20 = p2[0] - p0[0]
     dY20 = p2[1] - p0[1]
     dX10 = p1[0] - p0[0]
     dY10 = p1[1] - p0[1]

     s_p = (dY20*dX) - (dX20*dY)
     t_p = (dX10*dY) - (dY10*dX)
     D = (dX10*dY20) - (dY10*dX20)

     if D > 0:
         return (  (s_p >= 0) and (t_p >= 0) and (s_p + t_p) <= D  )
     else:
         return (  (s_p <= 0) and (t_p <= 0) and (s_p + t_p) >= D  )

你可以用:

X_size = 64
Y_size = 64
ax_x = np.arange(X_size).astype(np.float32)
ax_y = np.arange(Y_size).astype(np.float32)
coords=np.meshgrid(ax_x,ax_y)
points_unif = (coords[0].reshape(X_size*Y_size,),coords[1].reshape(X_size*Y_size,))
p_test = np.array([0 , 0])
p0 = np.array([22 , 8]) 
p1 = np.array([12 , 55]) 
p2 = np.array([7 , 19]) 
fig = plt.figure(dpi=300)
for i in range(0,X_size*Y_size):
    p_test[0] = points_unif[0][i]
    p_test[1] = points_unif[1][i]
    if ptInTriang(p_test, p0, p1, p2):
        plt.plot(p_test[0], p_test[1], '.g')
    else:
        plt.plot(p_test[0], p_test[1], '.r')

绘制网格需要花费很多时间，但是该网格在0.0195319652557秒内测试，而开发人员代码为0.0844349861145秒。

最后是代码注释:

# Using barycentric coordintes, any point inside can be described as:
# X = p0.x * r + p1.x * s + p2.x * t
# Y = p0.y * r + p1.y * s + p2.y * t
# with:
# r + s + t = 1  and 0 < r,s,t < 1
# then: r = 1 - s - t
# and then:
# X = p0.x * (1 - s - t) + p1.x * s + p2.x * t
# Y = p0.y * (1 - s - t) + p1.y * s + p2.y * t
#
# X = p0.x + (p1.x-p0.x) * s + (p2.x-p0.x) * t
# Y = p0.y + (p1.y-p0.y) * s + (p2.y-p0.y) * t
#
# X - p0.x = (p1.x-p0.x) * s + (p2.x-p0.x) * t
# Y - p0.y = (p1.y-p0.y) * s + (p2.y-p0.y) * t
#
# we have to solve:
#
# [ X - p0.x ] = [(p1.x-p0.x)   (p2.x-p0.x)] * [ s ]
# [ Y - p0.Y ]   [(p1.y-p0.y)   (p2.y-p0.y)]   [ t ]
#
# ---> b = A*x ; ---> x = A^-1 * b
# 
# [ s ] =   A^-1  * [ X - p0.x ]
# [ t ]             [ Y - p0.Y ]
#
# A^-1 = 1/D * adj(A)
#
# The adjugate of A:
#
# adj(A)   =   [(p2.y-p0.y)   -(p2.x-p0.x)]
#              [-(p1.y-p0.y)   (p1.x-p0.x)]
#
# The determinant of A:
#
# D = (p1.x-p0.x)*(p2.y-p0.y) - (p1.y-p0.y)*(p2.x-p0.x)
#
# Then:
#
# s_p = { (p2.y-p0.y)*(X - p0.x) - (p2.x-p0.x)*(Y - p0.Y) }
# t_p = { (p1.x-p0.x)*(Y - p0.Y) - (p1.y-p0.y)*(X - p0.x) }
#
# s = s_p / D
# t = t_p / D
#
# Recovering r:
#
# r = 1 - (s_p + t_p)/D
#
# Since we only want to know if it is insidem not the barycentric coordinate:
#
# 0 < 1 - (s_p + t_p)/D < 1
# 0 < (s_p + t_p)/D < 1
# 0 < (s_p + t_p) < D
#
# The condition is:
# if D > 0:
#     s_p > 0 and t_p > 0 and (s_p + t_p) < D
# else:
#     s_p < 0 and t_p < 0 and (s_p + t_p) > D
#
# s_p = { dY20*dX - dX20*dY }
# t_p = { dX10*dY - dY10*dX }
# D = dX10*dY20 - dY10*dX20

有一些恼人的边条件，即一个点恰好在两个相邻三角形的公共边上。这个点不可能在两个三角形中，也不可能不在两个三角形中。你需要一种任意但一致的方式来分配点。例如，画一条横线穿过这个点。如果这条直线与三角形的另一边在右侧相交，则该点被视为在三角形内。如果交点在左边，则该点在外面。

如果该点所在的直线是水平的，则使用above/below。

如果该点位于多个三角形的公共顶点上，则使用该点与中心点形成的角最小的三角形。

更有趣的是:三个点可以在一条直线上(零度)，例如(0,0)-(0,10)-(0,5)。在三角剖分算法中，“耳朵”(0,10)必须被切掉，生成的“三角形”是直线的退化情况。