These days, point clouds have become such an important tool in areas like geographical mapping, CAD and BIM modeling, etc, that it is worth writing some dedicated blog posts about them. This series of blog posts will cover topics like what point clouds are, the principles of scanners used to create them and their application in different domains.
A point cloud is nothing more than a collection of millions (sometimes billions) of points coming from a scanner. It is important to realize that these points are always located on the surfaces of objects. Each point has three coordinates to position it in space and often also color and/or intensity information.
You can watch point clouds in action at Bricsys 2018 here.
In a nutshell
One source of point clouds is laser scanners. This type of scanner creates a point cloud by emitting a laser beam in a certain direction (described by angles φ and θ). This beam reflects somewhere on a surface and the distance r to this reflection is measured. The result is one point in the point cloud (Figure 1 a). Sweeping this beam around and measuring the distances to all these reflections on surfaces results in a point cloud (Figure 1 b and c). Figure 1 d is then the ideal CAD model that we would like to reconstruct from the point cloud.
On the one hand, a point cloud is the most complete set of raw measurements of real-world objects possible. On the other hand, it is just a collection of “dumb” points without interpretation nor physical meaning of what is represented. Humans can easily recognize objects in point clouds, but they would really need a lot of time to do it for all meaningful objects in the point cloud. That is where clever software using Artificial Intelligence / Machine Learning can help to do part of this monotonous job for them.
Point clouds only contain information on the outside of objects. This is in contrast with data from ultrasound, CT or MRI scanners, which contain data of the full inside of the objects as well). Figure 2 gives a visualization of a point cloud of an elevator technical room. The same scene is shown in Figure 3 except that the front panels of the cabinets are now clipped away, showing that the point cloud has no information inside these cabinets.
Figure 3: The same scene as in Figure 2, but now also clipping away the front of the technical cabinets.
Some examples of domains that use point clouds:
- Creating or updating geographical maps with height information of large areas scanned from airplanes or satellites
- Providing a reference to create CAD models from real-world objects.
- Automatic quality inspection by comparing a computer model with the reality:
- Follow the different steps during production.
- Inspect a finished product for errors.
- Check deformations or other wear over time.
Use case old style
Imagine you are an architect with a project to retrofit or refurbish an existing building. Obviously, you want to start with a model of the situation as it currently stands. If such a model is not available (like for most buildings built before 1980), you need to create this. So you need a lot of measurements for this building.
First you must measure reference points like room corners. Starting from these corners, you can create walls and floors. From those, you can create your building model. In the past, people needed to measure these reference points by hand. With these points, they could then build the CAD model. As you can imagine, this is extremely time-consuming and thus expensive. Also, if during modeling it appeared that some measurements were forgotten, they needed to be done again.
Point clouds to the rescue
Nowadays, laser scanners like the Leica BLK360 can automatically measure half a million points per second with an accuracy better than half a centimeter. The result is a set of millions (in some cases even billions) of measured points with color information of a scene, in just a fraction of the time that humans could do it. This gives a much more solid reference to start modeling from. As a bonus, you get a nice visual impression of that scene as well.
Some people confuse a point cloud with a 3D model. Remember, a point cloud contains just exact measurement points, but NO knowledge of what these points represent. This is in contrast with a 3D model like a BIM. Such a model contains geometric solid shapes with information; their materials, what role in the building these shapes have (wall, support column, etc), what relation they have to other parts of the building (to which other walls/floor/ceiling are they connected), …
A future post in this series will discuss how such a 3D model is created from a point cloud. For now, just remember that creating such a 3D model from a point cloud even takes a human expert a lot of time and effort.
This series of posts mentions examples of companies / devices / software packages. These examples are not meant as an evaluation or a recommendation by Bricsys, but rather as illustrations of what currently is available in this fascinating field, both in academic research as in commercial applications!