Targetless Registration and the Future of Point Cloud Software

Targetless Registration and the Future of Point Cloud Software

Posted by LSA on 24th Jun 2014

As a seller of and general proponent of reference spheres, we’ve been slated to write a bit on the developing Cloud to Cloud (C2C) registration for quite some time now. As the uproar about targetless registration seems to be reaching its fervor, here are LSA’s thought on targetless registration, reference targets, spheres, and control/survey points.

(To preempt the discussion; at this point in time, and into the near future, it’s far wiser to use spheres/targets than to rely on the software. This most likely won’t always be the case, but it happens to be the case now.)

The primary issue surround the C2C debate, and indeed the source of many frustrations in point cloud data processing, is the task of precisely aligning individual scans and clusters with each other, i.e. point cloud registration. As we all (hopefully!) know, the objective when registering scan data is to eliminate or at least mitigate error between scans; hence the original purpose of the laser scanner target.

To that end, the target accounts for a shortfall in both hardware and software sophistication; necessity dictates that they would not exist, should the case be otherwise.

Likewise, necessity will dictate when they are no longer required. To be a little less obtuse, the best advice on spheres and registration one could possibly have is to implement a workflow that works best for the particular object or project you happen to be scanning. For example, larger projects that require greater accuracy, such as industrial plants, benefit from the structure control points offer. Featureless areas can pose confusing scenarios for the current software algorithms, and again reference targets provide a nice structure to work within.

In fact, one of the more comforting aspects of using spherical targets is the notion of a “control point” – in that the software is calibrated to recognize the sphere as exactly 145mm in diameter. Purely as a concept, finding a way to define and contextualize what’s observed (by a machine, animal, human, and so on) or measured is far more important than process itself. This, in a sense, is what the C2C algorithm attempts to accomplish.

A caveat to eventual phasing out of target based registration is the issue of tying in C2C registered point clouds to known survey points. Either by using spheres and special prisms, or something along the lines of our prism-sphere, the fact of the matter still seems to be that an identifiable point (target) is still required. Certainly a topic to be following as newer software develops.

In summation: we think targets should still be used, increasingly on a “just in case” basis. The new software seems promising, and with some much needed features added in, like target and C2C registration (we’re looking at you, FARO Scene), the results will most likely speak for themselves.