Algorithms

Arttu Soininen

Software developer

Terrasolid Ltd

Low points

●

For finding bad points clearly below the ground

●

Usually run before ground classification

●

Simple routine, can find only clearest cases

●

Not iterative, you should run it at least 2-3 times

Low points – Single points

●

Classifies a point if all other points Within

search radius are at least More than above

the point

Within

More than

Low points – Groups of points

●

Physical features may cause multiple points

below the ground at the same location

●

Groups of points option finds groups of

points which are all lower than any of the

neighbouring points

Low points – Multiple runs

●

Multiple runs needed for locations with bad

points below the ground at multiple

elevation levels

●

First run finds only the lowest and regards

everything else as valid points

Below surface

●

For finding bad points below the ground

●

Has to be run after ground classification

●

Can find points a little (15cm) below ground

Below surface

●

For each point, finds 6-12 closest neighbouring

points

●

Fits a plane equation to the neighbouring points

●

Checks the elevation distance to the plane

●

Regards point as valid ground point unless it is

at least Tolerance below plane

Below surface

●

Computes standard deviation of the distances

between the neighbouring points and the fitted

plane

●

If point being checked is more than Limit *

standard deviation below the plane, it will be

classified as a point below the ground

Smoothing

●

Laser data is dense but has a level of noise

●

Produces a rough, noisy looking surface

●

Smoothing modifies elevations of laser points

if that produces a smooth surface locally

●

Smoothing keeps the original elevation of a

laser point if the location does not become

smooth

–

This helps to keep breakline changes intact

●

Iterative process

–

During each iteration round, a point moves up or

down to match a plane equation fitted to

neighbours

Smoothing

original laser points

Smoothing

●

For each point, finds 6-12 closest neighbouring

points

●

Fits a plane equation to the neighbouring points

●

Closest neighbours have bigger weight in fitting

●

Point gets an elevation correction towards the

plane equation

Smoothing

●

During one iteration round:

●

Fits a plane equation to neighbours of every

point

●

Computes an elevation correction for every point

●

Applies the correction but limits the change from

the original elevation to be within maximum

movement setting

Smoothing

●

As a result of one iteration round:

●

Points match neighbours a little better

●

Iteration continues as long as there is movement

Smoothing

●

As a result of iteration:

●

Iteration stops when there is no movement

●

Some points do not match neighbours even

though they have been moved by the maximum

change

●

Routine restores the original elevation of those

points and their neighbours

Smoothing

●

Second iteration:

●

'Rough' points restored to original elevation stay

fixed

●

Second iteration is performed in order to make

'smooth' points match their 'rough' neighbours