Paul Mann

Overview

Background Outline of principles of methods Limitations Thoughts on design Potential outputs Variations Any suggestions?

BackgroundLast years symposium...

I liked Kevin Dixon’s laser scanner, a toy I could get used to playing with.

But most impressed by the potential of Anna Mason’s videogrammetry work.

VideogrammetryThe low contrast problem...

The method relies on tracking automatically identifiable features through a sequence of frames.Cave walls are often low contrast dull browns, lacking identifiable features.Shadows are the most readily identified feature, but these change position as the light source moves

Suggested solution...Add targets

But this takes time and is invasive.

So why not use projected spots of light?

Thinking time...

Trying out videogrammetryFreeware and demo versions of videogrammetry

software on internet.

Attempt to project light patterns

Also enhance edge detection

Plenty of difficulties, not many results

Sidelined“to build your own laser scanner for under a fiver...”

1. A computer2. A web cam3. A few sheets of paper4. A laser5. A glass fuse6. David...

Success at last!

How it works... Orthogonal background planes Marked with a calibration grid

Locates camera Scales and orientates planes Provides corrections for lens

distortions Structured light

Planar laser Calculate plane of light

From incidence on orthogonal planes Any other point lit must be on this

plane And on linear ray path calculated for

that pixel Hence point co-ordinates given by

intersection of these two.Image from David-Laserscanner .com home page

Drawbacks

The David system is constrained by its need for two reference planes.

It is possible to scan though a hole to capture detail a behind the planes, but range is very limited as laser line must still intersect both planes.

Like most structured light solutions it is “inward” looking.

Useful in caves?

Excellent for capturing details on a cave wallRock artScallops

But can we capture the whole cave?Back to the drawing board

More internet research...

Excel 2009 survey trade show in York Surveying is fun...

...it has lots of nice toys...

...but better to do it properly So off to Glasgow...

... project required

One solution

A structured light triangulation profiler

One solution

A structured light triangulation profiler

One solution

A structured light triangulation scanner

Design

The components:A planar laserA digital cameraA suitably wide lens

○ A parabolic reflector lens is a bit pricey, but ideal.

Fortunately I have a couple spare.

How it works

Photo OUCC archives: F2, The Font Pitch, 1983

How it works. Centre point equivalent to

down Fixed radius out is horizontal

ie a circle

Lens tangent point at edge of circleAbout 70 degrees upwards

How it works..

Vertical angle (clino) is a function of distance from central point

Horizontal angle (compass) is directly equivalent to angle from central point

How it works...

Laser sheet is perpendicular to camera axis, and offset from lens by a measured distance.

School geometry

Tan Φ = opposite

adjacent

Φadjacent =

offset

x

opposite = L, the distance from axis to point

Processing First, find illuminated pixels Each illuminated pixel has (X,Y) value

Convert this to image polar co-ordinates (r,ϴ) Convert image polar co-ordinates to

instrument space polar co-ordinates (L, ϴ) ϴ remains same Φ is function of r, L is a runction of Φ

ie L is a function of r Convert instrument polar co-ordinates into

real space co-ordinates (x,y,z) Need to know orientation and position of

instrument.

Complications to processing Barrelling / pincushioning of camera lens

Easily modelled Modelling of parabolic lens curvature

Again relatively easily modelled Offset between axis of two lenses

A bit more challenging

Either rigorous calibration to determine lens constants

Or skip and calibrate straight to a (X,Y) look up table

Limitations

PrecisionLimited by pixel spacingFor 6MPixel camera:

1000 pixels cover 160 degrees vertically

1 pixel equates to about 1/6 degree

For laser offset of 1m, and passage radius 1m this equates to 3mm precision

But because of the tan Φ this drops off rapidly as passage radius increases.

Limitations

Is the drop off in precision a problem in cave survey situations?I suspect notIn big passages, increase offset of laser

Need for low lightHard to make system work in daylightSuits cave survey

Output Remember, not just one point being

recorded –Typically a passage profile of 3000 pointsCaptured at reshoot rate of cameraCan easily distinguish red and green lasersCan easily determine above and below reflector

planeHence could create upwards of 10,000 points

per shotComparable to commercially available scanners

Challenges

Getting from a profiler to a scannerMovement need not be rotational, well suited for

linear capture

But how do we control such motion and record it accurately?

Stringing taut wires a lot of work

Shafts a lot easier to constrain

Possibly active railway tunnels (or even recording vegetation overhanging lines)

Handling data Pointclouds

A set of points defined by their x,y,z coordinates

May have other attributes linked (eg RGB)

Huge data sets Specialist software Getting away from paper

Variations

Other structured light solutions exist How about combining this approach with

photogrammetry / videogrammetry? Increasing computer power is making

complex solutions viable Robotics is driving a lot of research in

this area – worth keeping an eye on these developments

Over to you...

Thoughts about how this could be used, developed, etc very welcome.