image based scanning for buildings recording…. experiments in three scales

P1150037The other day during my very first ever really scary hard drive failure l lost some TST data from a survey of  a church in York where i have been doing some detail drawings for a project focussed on the north and west elevations.  I managed to recover much of the west elevation, but for the north i was facing a total resurvey.  Fast forward to me biting the bullet and finally getting around to a systematic field test of software based photogrammetry alternatives to the usual combo of terrestrial total station, hand drawings and laser scanner. Of particular interest here to me is the issue of scalability…

Below are three rather random examples but ones which serve the purpose – two on a macro scale (+/- 100m), but presenting very different geometry – and one of a micro scale on the order of only a few cemtimeteres in extent. We have a couple of different options for this kind of thing in our department including an old copy of  PhotoModeler and Topcon’s PI3000, but in this instance i used Agisoft Photoscan

York baptist church 1868

York baptist church: built 1868

Stage 1 In this first instance the task was simply to produce a conventional (and accurate to +/- 50 mm) set of elevation drawings for this mid 19th C church in short order, so nothing out of the ordinary.

This is an aisled church with aspirations towards medieval adornment in a rather restrained Victorian style. It is mostly constructed of West Yorkshire sandstone stained by atmospheric pollution for the better part of a century and a half of urban life. The geometry of the church is relatively complex with neo-gothic tracery, sculpted buttresses and multiple different building planes. I did a couple of runs at it  photographing the north elevation, backlit by a winter sun. In the first instance i took a limited number of photos from different angles. In the second, i went crazy with lots of photos, details and distance shots (with a lumix px3 with 24mm wide angle)

First attempt at data capture on north elevation: shows the area of interest with derived tie points from the combined geometry of 8 photos .

First attempt at data capture on north elevation: shows the area of interest with derived tie points from the combined geometry of 10 photos . Note that the surface has not yet been determined. The points are those common to photo pairs


Second screen shot showing the surface generated from triangulation of all relevant tied photos using software default (mesh generation smooth) for mesh resolution. Using a limited number of photos this produced a highly accurate scan of the N elevation, sufficient for the most part for generating accurate measurements of details and profiles Note that in area of architectural detail remote from the camera location the occasional data error occurs with a bubble or depression in the surface.


Photo-rendered surface with roof covers excluded. This was used to generate a metrically accurate elevation image for the north facade which in turn formed the basis of the drawn record. (spatially tied in limitedusing TST survey points)

York baptist church: second pass

Using significant more photos of both north and west elevations  the ‘scan’  becomes more complete. However this requires significantly more processing time especially to generate the kind of accuracy that i was aiming for, but i went with a the quicker less accurate processing options all the same. Result: more data errors in areas of challenging surface geometry. Decent all the same though.

Tie points generated from photo pairs of a much larger number of photos than i the first run (this time 20). Sloppiness on my part in positions the cameras evenly has generated some lighting and parallax errors which have generated significantly more data errors in the generated point cloud and surface. see next image

Surface model generated using standard default (not high accuracy) opotions for mesh generation. some surface errors in complex geometry as on the roof pitches and above the upper- left lancet window.

Fast forward: part 2

pyramid 1

Pyramid monument at Castle Howard built in1728 to designs by Nicholas Hawksmoor, standing on the line of the stray walls and referencing ancient examples, particularly the Pyramid of Cestius on the fortification of Rome

The other day Nigel and i were over at Castle Howard looking at revising a specification for repair of the pyramid monument on the outer faux fortifications of the estate. I took a run of photos around the monument for record purposes and later decided to put them though a scan test. Such a strongly geometrical structure situated on a base of complex rusticated decorative masonry presented a great opportunity to test the boundaries of the method with some bad photos and what could be made of them.

One of the notable results of this test was that even using a limited amount of photos, the masonry of the base modelled extremely well given the detail on the photos (would probably have been better had I allowed the highest accuracy settings). The smooth surface of the pyramid itself, which is built from well dressed West Yorks sandstone ashlar blocks scanned well, particularly where there were fractures in the stone from rust-heave of the iron clamps. The smother surfaces created some difficulties however, with repetitive ridges probably resulting from parallax errors from the limited number of cameral angles.


Tie points from photo alignment. Note here the lower density on the shadowed sides of the Pyramid. Photoscan benefits greatly form well lit photograph it seems


Triangulated irregular Network (TIN) from first scan of surface showing actual geometry inferred from photo pairs. Accuracy levels are very high here, though on the smooth surface of the pyramid itself there appears to be a resonating data striping. I think this results form parallax problems with my photo angles, together with the available light.


Surface model of the same showing wave patter onthe pyramid, but very good detection of masonry defects from the spalling from iron cramps (usually saplling the top of of the stone on which the clamp is placed, with the upper stone undamaged). The masonry of the plinth scans exceptionally well.

Stepping down: part 3

joint 2Pushing this to really small scale targets, today i got around to doing some tests on the joint carpentry of a floor beam that i recovered from the skip during works to the Kings Manor 10 yesrs ago (dont worry, the University have mostly stopped ripping fabric like this out of their grade 1 listed buildings :-).

I did two tests.  One of a mortice for a floor joist with its peg still in place, chisel marks from the cutting of the mortice, and hewing scars. The second, of the tenon on the end of the beam which has very faint scribing marks illustrating the ‘double cut’ method that was used to layout and cut the joint.

With the first example i used  three photographs focussed on the the mortice for the floor joist (c.75mm wide and 100mm high). The aim here, to test a limited photo range’s capabilities for scanning the strong geometry of the joint together with the more irregular aspects of the grain structure of the oak, peg, and associated tooling marks from hewing axe and chisel. The outcome was pretty good, with some surprising results including the detection and modelling of the occasional cobweb on the surface of the timber. The geometry of the joint came out very well with even very fain scaring of the timber from the chisel cut rendering out accurately


Surface model of an area of timber c.200mm across. The joint itself is 75mm x 100mm with a sloping ‘throat?’ with dept of 30mm. The Peg is still in . the joint


Detail of TIN surface in the joint.


Deail of surface model of the joint. Note the grain structure inside the joint itself, and particularly the scar of the chisel on the right hand shoulder where the carpenter has cut off the line.


Photo-rendered surface with highlighted circle to show contrast between surface and draped photo. The draping of the photo goes a long way to articulating detail on this kind of scan, though doe not add any detail tot he surface itself.

One thing that i was pleased by – because i have never been a big fan of photo-rendered models (I find them strangely unsettling in the way that they distort a real world sense of space ) – os the way the a photo texture enhances detail on this scale of object, making clear what is sometime ambiguous from surface alone. In fact in this instance, even though it is not actually doing so, the photo corrects some deficiencies in the scan.

Stepping right down

This speaks for itself really. Only thing to say is that i was trying to capture the scatchawl scribe marks dor the ‘double cut, and the level marks. it picked up the deeper race knife marks well and the hewn surface, but apart from a some fain traces of the scribe marks , these were a step to far.

Screen shot 2012-12-08 at 21.06.42

TIN model of the end of the same beam with assembly mark on the left face, chipped arrises near the joint itself and the shoulder of the joint (joint not modelled in detail)

Screen shot 2012-12-08 at 21.06.33

Surface view of the same showing grain of timber, assembly marks, and the scibe marks for the shoulder of the joint can just be made out (though the first scribe for the first cut which is only present on the side of the tenon did not sow up in the surface

Screen shot 2012-12-08 at 21.07.09

Photo-rendered surface

Screen shot 2012-12-08 at 21.08.13 Screen shot 2012-12-08 at 21.08.20BG 08-12-12

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s