Ejectamenta News Summary

Coordinate transformations in 3D

Tutorial using Matlab to show how to perform transforms in 3D from one coordinate system to another

For real world vision or Robotic operations it is sometimes necessary to transform between coordinates in different coordinate systems. In this tutorial we will transform a 3D spiral from one coordinate system to another. The coordinate systems are specified in terms of their position in Cartesian space and converted to a vectorial representation.

First let us create and display the spiral

t = 0:pi/50:10*pi;
st = sin(t);
ct = cos(t);

The spiral is in 3D but we need a 4D matrix to perform the matrix transformations, so we also add a row of ones to it


spiral = [st; ct; t; ones(501,1)']

Ok, lets plot the relevant data


plot3(spiral(1,:), spiral(2,:), spiral(3,:), 'b')

we use the hold on command to keep plotting to the same chart


hold on

We are going to translate this spiral from its world coordinates to another local coordinate system

the coordinates are specified by a matrix = [X1, X2, X3, Y1, Y2, Y3, Z1, Z2, Z3]


world = [1,0,0;0,1,0;0,0,1]
local = [3,6,3;3,3,6;3,3,0]

The 3,3,0 in the last row (Z axis) of the matrix creates a 45 degree plane

Lets turn our local coordinate system into a vector


local_vector = zeros(3,3)
local_vector(:,1) = local(:,2)-local(:,1)
local_vector(:,2) = local(:,3)-local(:,1)
local_vector(:,3) = cross(local_vector(:,1),local_vector(:,2))

The vectors should be normalised to unit length like our world vector is


local_vector(:,1) = local_vector(:,1)/norm(local_vector(:,1))
local_vector(:,2) = local_vector(:,2)/norm(local_vector(:,2))
local_vector(:,3) = local_vector(:,3)/norm(local_vector(:,3))

Now we have to create our transform matrix using the dot product between our world and local vectors, this sets up the skew, reflection, rotation elements


transform = zeros(4,4)
transform(1,1) = dot(world(1,:), local_vector(1,:))
transform(1,2) = dot(world(1,:), local_vector(2,:))
transform(1,3) = dot(world(1,:), local_vector(3,:))
transform(2,1) = dot(world(2,:), local_vector(1,:))
transform(2,2) = dot(world(2,:), local_vector(2,:))
transform(2,3) = dot(world(2,:), local_vector(3,:))
transform(3,1) = dot(world(3,:), local_vector(1,:))
transform(3,2) = dot(world(3,:), local_vector(2,:))
transform(3,3) = dot(world(3,:), local_vector(3,:))

we also add the translated origin of our coordinate system to the last column of the matrix, this performs the X,Y,Z translation


transform(1,4) = local(1,1)
transform(2,4) = local(2,1)
transform(3,4) = local(3,1)
transform(4,4) = 1

So using our transform we can rotate the spiral by 45 degrees and translate it by 3 in X and 3 in Y


tspiral = transform*spiral;

lets plot it in green next to the original spiral


plot3(tspiral(1,:), tspiral(2,:), tspiral(3,:), 'g')

(you might want to rotate the plot axis now to see the two spirals in 3D)

now to get it back again we use the inverse of our transformation matrix


tunspiralInv = inv(transform)*tspiral;

and plot it using markers so we can see the orginal spiral beneath


plot3(tunspiralInv(1,:), tunspiralInv(2,:), tunspiralInv(3,:), 'r*')

The plot should look like this

plot of spiral being translated between coordinate systems

For the mathematical details behind the transform, these MIT lecture notes provide an excellent explanation.

If you find the article useful, please feel free to post a link to it


Return
May 17, 2018 Category: maths Posted by: dave