3D Modeling - Point clouds vs meshes

Three-dimensional modeling is far from a new technology, but the methods and devices used to make 3D models are rapidly developing and improving. There are quite a few different technologies available to anyone with the interest (and budget) for capturing accurate digital models of the physical world, e.g. laser scanners, structured light scanners, LiDAR sensors, photogrammetry, etc. Each of these technologies have their own benefits and limitations (usually in the form of their scanning range or the environment they can fit or operate in) but they all have the same endgame - to make a point cloud.

There’s lots to say about these technologies but, for today, we’ll be focusing on understanding their end result and the two most commonly used terms to describe the end result: Point Cloud and Mesh.

Laser Scanner

Laser Scanners, like this Leica RTC360, are often mounted to tripods. These scanners can have large operating ranges (exceeding 100 yards) but still need to be moved around the site or object to capture every angle and detail.

Structured Light Scanner

Structured Light Scanners, like the Artec Eva shown here, work best for scanning small to medium sized objects. These scanners can fit into small spaces that most laser scanners can’t and often capture more detail at finer resolutions. Here, the structured light scanner is being used to capture a car’s crankshaft.

LiDAR

LiDAR sensors come in many forms. The senor shown here (made by FARO) is referred to as a “SLAM” type (Simultaneous Localization And Mapping). These sensors capture data on the move and excel at rapid capture in challenging conditions.

Photogrammetry

Photogrammetry is the art and science of capturing three-dimensional information from photographs. Any camera, from drones to cell phones, can be used to take the photos. However, the camera used (and skill of the photographer) can drastically affect the quality of the results.

Point Clouds

Whether you’re using a laser scanner, structured light scanner, LiDAR sensor, or photogrammetry to make a digital model, you’re going to be dealing with point clouds. Point clouds are literally a grouping (or “cloud”) of millions of individual “points” in three-dimensional space that describe the object they’re representing. The position of each one of the millions of points represents the actual position of a real location on the scanned object. Oversimplifying a bit, these points basically just trace an outline of the shape and size of the object. The figures below should help reinforce the concept.

A Pot and a Point Cloud

This piece of pottery is shown here in front of a point cloud of the same object. This point cloud was made from photographs of the pottery (photogrammetry).

Point Cloud

Notice how the pottery is see through? That’s because we’re looking at one million floating points that outline the shape of the pottery.

Solid Model

Skipping ahead a bit, here we see the finished model of the same piece of pottery. Notice how the model is now solid rather than see through.

Point clouds are arguably the most important part of a three-dimensional model when it comes to forensics. That’s in large part because the point cloud is essentially the raw data, or measurements, taken by the laser scanner (or whatever other device is being used). The point cloud is essentially the framework for the model and it determines the accuracy of any measurements taken from the 3D model.

Meshes

Point clouds form the outline of the object - Meshes connect the dots. Specifically, meshes connect the points to the other points around them using tiny, solid, triangles to fill the gaps between the points. You can think of a mesh as the skin that wraps around the point cloud - like draping a bed sheet over a birdcage.

Let’s look at some more examples. The figure below shows a 3D model of a construction site made using photos from a drone (photogrammetry again). From a zoomed-out perspective, you can’t see most of the gaps between the individual points that form the point cloud. As a result, the point cloud and mesh don’t look much different.

Point Cloud

Mesh

Colorized Mesh

Zooming in highlights the differences between point clouds and meshes. The figure below shows a dirt mound from the figure above (the one encircled by the road) in more detail and from a different perspective.

Point Cloud

After zooming-in, you can now see the individual points that make up the point cloud.

Colorized Mesh

The mesh filled the gaps in the point cloud to make a solid 3D model. Color and textures from the original photographs are projected onto the mesh to make this a colorized and textured mesh.

Let’s look at one more figure to really drive the point home. The figure below zooms in on a tree that was only partially captured by the drone photos (the tree is on the top left of the first figure from above). The leaves around the top of the tree were captured by the photos and are fairly well represented by the point cloud. However, the drone photos didn’t capture the lower portions of the tree - leaving large gaps in the point cloud. Nevertheless, the meshing process connected the dots and filled the large gaps in the point cloud, leading to a tree that looks more like a green blob.

Remember when I mentioned that point clouds are the most important part of the model for forensics? This example demonstrates why. The point cloud clearly shows large gaps in the data where no points were collected while the mesh covers up those gaps. Covering the gaps is desirable for a nice dense point cloud that accurately defines the shape, size, and contours of the object but it can also hide/obscure deficiencies in the model, resulting in a mesh that doesn’t accurately represent the shape, size, or contours of the object.

Point Cloud

The point cloud makes it clear that not enough data was captured to adequately model this tree. There simply aren’t enough points.

Mesh

Here’s the mesh before colorization and texturing. This image clearly shows the triangles that make up the mesh. Here, the triangles are huge because the gaps between the points they connect are huge. We want tiny triangles for a good mesh.

Colorized Mesh

Notice how the large triangles are now a blurry green color.. None of the drone photos captured that part of the tree so the program stretched the colors across the gap - leaving a blob that doens’t adequately represent the tree.

Final Thoughts

Today’s post is a 10,000ft view of 3D modeling and intentionally skips over important differences between the technologies used to make these models. For example, our first blog post discussed the importance of control points for establishing the accuracy and reliability of photogrammetric models made from drone images. That post nicely highlights why being able to use the equipment and make a model isn’t a substitute for understanding the theory behind the methodology and the limits of the technology.

We’ll be diving deeper into other 3D modeling technologies and use cases in future posts.