# Equidistant points on a polyline

Set of 2D-points connected by a polyline B-spline function:

p = RandomReal[{-1, 1}, {20, 2}];
f = BSplineFunction[p,
SplineDegree -> 1,
SplineClosed -> True];


This is neatly defined polyline function. However, as the spline parameter goes from 0 to 1 with some constant step, calculated points are somewhere denser than at other regions. (I can see the density decreases with distance between points.)

Graphics[{
Point[p], Opacity[.2],
Point[f /@ Range[0, 1, .001]]}]


Me I need a function that returns equidistant points for equidistant parameter values.

Graphics[{
Point[p], Opacity[.2],
Point[g /@ Range[0, 1, .001]]}]


Where g I constructed like this:

g[t_] := Evaluate[
With[{u = With[{
d = EuclideanDistance @@@ Partition[p, 2, 1, 1]},
Accumulate[d]/Total[d]]},
Piecewise[Table[{
p[[i]] + (t - If[i > 1, u[[i - 1]], 0])/
(u[[i]] - If[i > 1, u[[i - 1]], 0])*
(p[[If[i != Length@p, i + 1, 1]]] - p[[i]]),
t <= u[[i]]}, {i, Length@p}]]]]


This can be made more elegantly, right, with Mathematica? With some option that samples equidistant points? Because say I have some smooth curve function. I don't know how I would tackle this then. I guess one would have to integrate and findroot some.

• Do you actually need the values of the points, or do you just want to connect them with nicely spaced dotted lines? Commented Mar 24, 2013 at 16:57
• Related or duplicate? Generating evenly spaced points on a curve Commented Mar 24, 2013 at 17:15
• @bills Values, yes. Otherwise Dashing? :) Vitaliy linked to an elegant solution, will study that.
– BoLe
Commented Mar 24, 2013 at 18:02

p = RandomReal[{-1, 1}, {5, 2}];
f = BSplineFunction[p, SplineDegree -> 1, SplineClosed -> True];


A very simple approach:

np[u_, dt_] := u + dt/ Norm[D[f[t], t]] /. t -> u;
ListPlot[Table[f[t], {t, NestWhileList[np[#, .03] &, 0, # < 1 &]}], AspectRatio -> 1]


Testing that the points are equidistant:

ListLinePlot[EuclideanDistance @@@
Partition[Table[f[t], {t, NestWhileList[np[#, .03] &, 0, # < 1 &]}], 2, 1],
AxesOrigin -> {0, 0}]


The only small exceptions are at the original points, as expected.

Edit

For a higher degree interpolation:

p = RandomReal[{-1, 1}, {7, 2}];
f = BSplineFunction[p, SplineDegree -> 5, SplineClosed -> True];
GraphicsRow@{
ListPlot[Table[f[t], {t, NestWhileList[np[#, .003] &, 0, # < 1 &]}], AspectRatio -> 1],
ListLinePlot[EuclideanDistance @@@
Partition[Table[f[t], {t, NestWhileList[np[#, .003] &, 0, # < 1 &]}], 2, 1],
AxesOrigin -> {0, 0}, PlotRange -> All]}


• Basic-calculus approach, clean and tiny. Thank you. The jumping around the original point parameter values isn't noticeable in my animation -- and I can decrease the step size which diminishes the jumping, and then Take only every second, third or so point since there are probably too many for my purpose.
– BoLe
Commented Mar 25, 2013 at 12:57
• @BoLe Yep. The idea was to keep it simple :) Commented Mar 25, 2013 at 14:04

You need what is sometimes called a reparametrization by arc length. Since the velocity is piecewise constant, it might done as follows:

p = RandomReal[{-1, 1}, {20, 2}];
f = BSplineFunction[p, SplineDegree -> 1, SplineClosed -> True];
arclengths = Accumulate[Norm /@ Subtract @@@ Partition[p, 2, 1, 1]];
totalarclength = Last[arclengths];
t = Interpolation[
Transpose@{Prepend[arclengths, 0.], Range[0, 1, 1/Length[p]]},
InterpolationOrder -> 1];

Graphics[{Point[p], Opacity[.2],
Point[(f[t[#]]) & /@ Range[0, totalarclength, totalarclength/1000]]}]


Or one could use calculus to find t as in the link in Vitaliy Kaurov's comment that I just noticed, or in this alternative way:

v[t_?NumericQ] := Norm[f'[t]];
t = NDSolveValue[{tt'[s] == 1/v[tt[s]], tt[0] == 0}, tt, {s, 0, totalarclength}];