Cloud2cloud registration and registration with target marks
The FARO SCENE update 5.3, 5.4, and 5.5 means that target-less registration (also referred to as cloud2cloud registration) is a very current topic. The function, however, has already been on the market for a while. Various software packages of other laser scanner manufacturers such as Leica Cyclone register, Z+F LFM, and Riegl Riscan have included this or similar functions for some time now.
We often receive calls from customers who have naively deployed these new stationing possibilities and have then encountered problems registering scan data in FARO SCENE. The programs of the manufacturers are not to blame here; the problem lies in their deployment by the user for stationing point clouds.
For example, many Leica customers continue to use target-based solutions for the stationing of laser scans. There are also fully developed functions of this nature in other areas of 3D point cloud software packages, including the programs of 3D Systems (Geomagic Wrap, Control, and Design X) and Polyworks. These approaches are also used in industrial measuring technology. Target-less registration (cloud2cloud registration) is not so well recognized here. It often lags behind the "Best Fit" function, for example. With target-less registration, the calculation involves points, areas or normals. The most important fact is that how well the different scan data records overlap is decisive for the later quality of the registration.
An example from practical application
If a number of scans are carried out in an enclosed room, target-less registration is very exact, because in this case the calculation involves millions of identical points. This creates a very precise calculation of the average and the scans can be calculated very well in relation to one another. Another important point is covered very well here. In order to be able to calculate a target-based registration (cloud2cloud registration) in the first place, the algorithm requires a stable source value of the scans in relation to one another. In the case of scans in a room, this is very simple. If, on the other hand, the scans are of a large building such as a school, it becomes very difficult for the algorithm also to specify the starting values of its calculation. This is especially the case when different room geometries are repeated and/or also occur in different floors. This means that there are multiple identical combinations here that match each other geometrically. For this, FARO SCENE includes the 'top-view-based registration' function, i.e. the user shifts the scans in the software onto the best starting position if the SCENE software does not find it automatically. What is interesting is whether this only works well with a single scan or also with many scans.
On the subject of overlaps as the area that is identical between 2 scans: the larger the overlapping area, the better the expected result. If the scan recording moves through the building, there will also be areas where the overlapping areas are rather unfavorable, e.g. in stairways or passing through small access rooms. Here, with unfavorable or inadequate overlap, the quality of the registration can drop and/or solution of the transformation might not be possible, i.e. the user does not receive a result. With most registration algorithms, it is also the case that only the neighboring relations between 2 scans are examined, i.e. you move hand over hand with your registration through the building. A poorly registered scan means that all scans attached to it are then shifted by this amount in the overall model.
Outdoors, a target-less registration is often not possible, as here there are only points on the ground and without constructions the scans cannot be supported at the side. In addition, the ground points are often the ground vegetation, i.e. the identical points are more diffuse than correct. The algorithm often fails here. Especially outdoors, targets such as reference spheres or checkerboard markings are a significant advantage. In the case of target-less registration in FARO SCENE 5.3 and higher versions, all measured points are used as a general principle. Outdoors, this can be passing vehicles and/or the vegetation that changes with the wind. It can also involve ghost points of multiple measurements on window panes and other optical effects. This also applies to indoor measurements, i.e. in the case of target-less registration as in FARO SCENE, the work is performed with measuring points that are known not to be correct and this is relied on, i.e. the number of good measuring points is significantly better, which makes the weighting better. For users who have zero tolerance in relation to measurement errors, it is better to use targets here from the outset.
Another important point is the computing time. During target-less registration, a great many measuring points are calculated together. In the process, manufacturers have been known to use tricks: namely that not all point are used in the calculation, rather there is a subsample, i.e. target-less registration calculates with thinned-out point clouds, which means that it is incorrect to say that all points are used. Whether the algorithm of the cloud2cloud registration is intelligent is unknown. Another approach is that areas are fitted beforehand and then only the area is calculated together. Area fitting takes a very long time; calculation via the identical areas in target-less registration is relatively fast, although this is not very much different from a registration using target marks.
If a cloud2cloud registration is calculated via the point cloud, this can take a very long time. It is our experience that the algorithms are also apt to crash occasionally with large scan data records and that no result is achieved after a great deal of computing time.
There is also only one specification of the accuracy achieved by the registration as a mean value. The precise computing processes are not easy for the user to follow. It is a black box algorithm.
When can target-less registration be used?
In small indoor areas, target-less registration can be a good alternative to registration with targets. In doing so, the number of scans should not be too high and there should always be good overlaps between the scans. Linear objects such as long hallways or roads should not be scanned in this way, as error propagation occurs between each scan. For many scanners, the average measuring accuracy is approx. 2 mm per scan, i.e. with only 10 scans, there can be an error of more than 2 cm.
A solution can also be a mixed registration with cloud2cloud and target-based registration.
When can target-less registration not be used?
Definitely not outdoors if there is not much construction present. There might simply not be enough good overlapping areas. In addition, projects with a high number of scan locations are not recommended either. The risks of recurring geometries, poor overlapping areas, and vast but insoluble computing time are too great.
In general, this is also the case of you want to be sure when leaving the scan location that you will not have any nasty surprises later on registration. Pass marks well placed by a trained operator are always a guarantee of a good registration result.
When verification of accuracies is required, a cloud2cloud solution is not recommended either. Here, the person performing the calculation can have difficulty verifying the accuracy, and the repeatability of results is also very difficult. In the event that the measurement results are called into question, there might be some serious difficulty in providing explanations. This is why this registration is prohibited for many applications by standardized work instructions in metrology (industrial measuring technology). Also in the area of hand-held scanners, it is often required to deploy additional targets if an object geometry does not permit adequately unique overlapping areas.
An interesting new approach here is also the geodetic registration of the software company PointCab, which is available as a plug-in for FARO SCENE. This registration software offers for the first time the opportunity to specify a measurement report for the FARO SCENE registration which reflects technical measurement standards.
Conclusion regarding target-less registration in comparison with registration via targets
The new registration methods in FARO SCENE – cloud2cloud and top-view-based registration – are additional interesting tools that can reduce the overhead of a well-versed user. A requirement here, however, is that the user understands the material, whereby a training course can be very helpful.
How can we at Laserscanning Europe Gmbh help you?
We provide regular training courses on the subject of laser scanners + FARO SCENE, including training for registration such as cloud2cloud registration, top-view-based registration, and registration with targets and measurement coordinates.
Over and above this, our laser scanner web shop offers a large number of different pass marks such as reference spheres, easy targets, and checkerboard targets for a wide variety of applications.
Author: Dipl.-Ing.(FH) Eric Bergholz, Laserscanning Europe GmbH
Co-author: Dr.-Ing. Richard Steffen, PointCab Software GmbH
We would be glad to receive suggestions, answer any questions, and hear about new working methods regarding the problems of stationing scan data.