[x,y,z] = CreateUniformDotsIn3DFrustum(ndots,FOV,aspectr,depthrangen,depthrangef,eyeheight)

Sample dots in frustum uniformly, creating a cloud tightly fitting in the
frustum. When the optional 6th parameter eyeheight is given, dots will be
uniformly sampled on a ground plane at -eyeheight.

z is not sampled from a uniform distribution, but from a parabolic
distribution as the area of cross sections of the frustum is a quadratic
function of the depth plane’s depth ( (z*2*tan(FOV/2))^2 * aspectr )

Here, I use Inverse transform sampling to transform a uniform random
variable into the quadratic shape random variable
see Luc Devroye. Non-Uniform Random Variate Generation. New York:
Springer-Verlag, 1986. Chapter 2

compile following in latex to see full derivation

\usepackage{amsmath} % math

Use pdf related to cross-section surface of frustum:
(\left(2 z \tan\left(\frac{FOV}{2}\right)\right)^2 aspectr)\
(z_1) is the distance of the near depth plane\
(z_2) is the distance of the far depth plane\
(y) is a uniform random variable\
Given: (F(z_2)=1) and (F(z_1)=0).
F(z) &= \int\limits^z_{z_1} k z^2 \, \mathrm{d}z\
F(z) &= \frac{k}{3} \left(z^3 - z_1^3\right)\
k &= \frac{3}{z_2^3-z_1^3}\
F(z) &= \frac{z^3-z_1^3}{z_2^3-z_1^3}
Substitute (y) for (F(z)) and factor out (z):
z^3 &= y\left(z_2^3-z_1^3\right) + z_1^3\
z &= \sqrt[3]{y\left(z_2^3-z_1^3\right) + z_1^3}
For a ground plane, the width of the frustum at depth (z) is given by
(2 z \tan\left(\frac{FOV}{2}\right) aspectr).\
By following the same inverse transform sampling steps, we would end up
[ z = \sqrt{y\left(z_2^2-z_1^2\right) + z_1^2} ]
for the linear distribution between (z_1) and (z_2).


Path   Retrieve current version from GitHub | View changelog