We noticed an interesting trend in responses to a recent blog post, “The Next Evolution in Surveying .. High Definition” .. quite a number of engineers got in touch with questions about scanning and point clouds. We thought we should do a follow up post specifically addressing some of the queries with a specific focus on how the engineering profession might leverage point cloud data.
In this blog post:
- What is a point cloud?
- How is the data captured?
- What can you expect from a point cloud?
- What sort of accuracies are we talking?
- What software can leverage the data?
- What can be modelled out?
A point cloud is a set of data points in some coordinate system. In a three-dimensional coordinate system, these points are usually defined by X, Y, and Z coordinates, and often are intended to represent the external surface of an object. Point clouds may be created by 3D scanners. These devices measure a large number of points on an object’s surface, and often output a point cloud as a data file. The point cloud represents the set of points that the device has measured.
The laser scanner is set up on a tripod, same as a total station. It is not a whole lot bigger either. Special targets used in high definition surveys are deployed around the site to assist with coordinate control.
We typically deploy a combination of a total station (theodolite) and a laser scanner to a site. The “traditional” total station assists with the survey control – it allows us to get the point cloud accurately on to the same coordinate system / site grid and height datum (AHD say) that you are using with all of your other data: civil, topo, Revit, aerial photography and so on.
This hybrid deployment, which only surveyors will offer, allows for flexible deliverables. Aside from the point cloud, traditional CAD files derived from a total station are a highly efficient way to describe certain types of features at a site – you still need to use the right tool for the right job.
Where a site has survey control, our particular choice of a “survey-grade” laser scanner allows us to mesh the traditional CAD files and laser scan point cloud data with impressive accuracy. Just as quickly, our laser scanner has a dual-axis compensator – same as our total stations (it is a levelling thing). It is also the same laser. This means the measuring specifications of the scanner and the total station are near on-par – these instruments are no toys! But they are pretty cool ..
The laser scanner pulses a laser at ~ 50,000 times a second. The laser reflects off surfaces it hits out to about 300m away and the scanner can “see” this reflection. The time-of-flight for each pulse is measured and therefore the distance to the surface is known.
The scanner also rotates through the horizontal axis during the scan, while the scanner head spins through the vertical axis at high speed. This now means we know the direction to the reflected surface relative to all other measurements from that location – each reflective laser pulse has an x,y,z coordinate and also the intensity of the laser return. All of this data is written to the internal solid state hard drive for safe keeping.
When visualised in a graphics package – this database of reflective laser pulses resembles a cloud.
When the scan at the current position is completed, we can also instruct the scanner to take up to 270 high resolution photos. The photos are a further record of the site. The photos can also be used in an additional process back in the office to apply colour to the point cloud – so that the points of the cloud in the graphics software can take on the colour of the surface the laser hit. You can easily alternate the view of the point cloud between true colour, orange (say) – scaled by the laser’s return intensity, black and white or other schemes – one popular one is colours by point elevation or “normals” – based on perpendicular vectors to the surface.
The scanner’s “resolution” (per scan) is configurable – a single scan take from a minute or two, up to a couple of hours – depending on the level of detail required – the maximum resolution of our instrument is described as the distance between points measured on a grid at 100m away from the scanner – a point every 20mm. Because the scanner measures in arcs out from the position of the scanner, the resolution rapidly increases the closer you get to the scanner – so at this resolution at 10m from the scanner you are talking a point every 2mm.
You typically achieve even higher point densities though! When you move the scanner to a new location and scan again, you will have overlap – so a pipe or a column say will be scanned from several locations – meaning more and more measurements covering the surface of the column.
Lasers can only measure line of sight – so by repeatedly moving the scanner you fill in the “shadows”. To get four sides of a column you will need at least two scan stations for example.
This does introduce a complication – all the scans need to be joined up – all put on the same coordinate system. This process is called registration. By registering all the point clouds for a site, you can then visualise the clouds simultaneously – you can then orbit around the column and view it from all sides or interrogate the cloud – for example, determine the size of the column. This is why survey control and proper field methodology is critical – this overlap. If the final derived coordinates for a station or stations are wrong, even by a centimetre, the surfaces will not match perfectly across multiple scans. Conveniently the Leica Cyclone software we have goes some way to mitigate this effect through what is called cloud-to-cloud registration – the software automatically compares millions of points to tighten the overlaps. Notwithstanding – we prefer to see the superb results we get from deploying the total station and scanner together. We typically get better than half-centimetre comparisons between control targets over kilometre ranges.
So when we are talking pricing and deliverables – one of our key products is a Registered Point Cloud – meaning the scanning of the site, all scans subsequently joined together on a coordinate system. Additional variables include level of detail required, survey control and photos for colouring.
You can read all about the accuracies of the relevant scanner – and please ask if you want to know specifics – but simply put, they are pretty bloody good.
Each point in the cloud is good to a few millimetres relevant to the position of the scanner. The position of the scanner in each scan can be coordinated between scans to within a few millimetres. This all means we can easily achieve tolerances requested in typical tenders – 20mm absolute and 10mm relative. We do suggest though for specific tasks – maybe if retrofitting an expensive steel beam say or where tolerances are significantly tighter, we double check any critical measurements with a total station.
When reviewing a “surface” (say a face of a column) – it is usually several millimetres “thick” – meaning the real answer as to where the true surface lies is an average of all of those points. Our software modelling tools take this into account when “growing” or extracting pipes, patches, boxes and other geometric shapes.
It is interesting to compare the point cloud deliverable to the traditional total station pick up.
Taking a simple example – points on a ground/surface grid – the surveyor using a total station might pick up a point on a slab 10 metres apart (say) – deploying the scanner means a point is available every couple of millimetres. Software algorithms can take the huge number of points and offer points on a grid based on options like “lowest point” or “ground point”.
Likewise – consider a concrete beam – our surveyors traditionally would pick points on the beam they think are representative of the shape of the beam, you might get some line work in a CAD file joining half a dozen points describing the beam – now compare this to the same beam defined by literally hundreds of thousands of points captured in a point cloud. Exciting times indeed!
At their simplest – point clouds are simply a database of points – x,y,z coordinates. This makes the data extremely portable. We can export point clouds in file formats such as plain-old text files, DXF’s, binary/compressed formats and more.
Looking to more proprietary formats, we can distribute files in *.rcp format – as in Autodesk ReCap files. This means files can be loaded in 2014/5 versions of Autodesk Revit, Civil3D, Navisworks and more. This means you can visualise current, as-constructed data, right along side you proposed models.
Navisworks Manage (amongst other packages) can do more advanced things like clash detection. You can load up existing structures capture as a point cloud and compare them to proposed models to ensure the new designs will fit.
Point clouds are great for ad-hoc interrogation, visualisation and animation and clash detection workflows. Invariably though, users will want CAD style outputs derived form the point cloud.
We can create models and line work from the point cloud using specialised software. It does take time to do – you can liken the software to the early days of OCR (optical character recognition). As the software matures, more and more automated feature extraction functionality is added. We can currently extract things like: pipes; surfaces (patches); steel work (like “I” beams) based on library models; meshes; and ground points-on-grids and contours – but it is a fairly labour intensive process.
These features can then be moved around as Autodesk faces and solids for example, or imported into Revit as masses.
Revit is a little trickier – it likes things to be perfect – floors to be perfectly flat, walls to be vertical or right-angled, columns all to be precisely square and same uniform size. Buildings are never built perfectly – and the point cloud shows it up! That is not to say you can’t model in Revit – but we will have a discussion with clients based on the level of detail (LOD) required. For example – does Revit need to know a wall is built at 87.2 degrees vertical, or can the wall be approximated in Revit. For a building information model (BIM) – this is probably all that is required – but if you wanted to retrofit an expensive steel beam – you would refer back to the original point cloud for more precise data about the features.