July 29, 2019

Introduction to 3D point clouds

What is a 3D Point Cloud ?

A 3D point cloud is a list of cartesian coordinates X, Y and Z representing the geometry of a scene. 

Generally composed of millions or billions of points, they are accurate digital representations of the real world. These coordinates correspond to the actual size of objects. Thus, we can make measurements, inspections and inventories as if we were in the field.

3D Point clouds are usually acquired by LIDAR scanners installed in terrestrial or aerial vehicles. The geo-referencing is usually carried out thanks to Global Positioning Systems (GPS) and Inertial Measurement Units (IMU). Depending on the acquisition system, some additional attributes such as laser intensity, GPS time or color may be available.

This technology is particularly well suited for mapping large scale infrastructures such as electrical corridors, railways, highways, urban environments, among others.

Example of Mobile Laser Scanning (MLS)

What is 3D point cloud classification ?

Classification is the process of giving a semantic sense to each group of points in the 3D point cloud. For example, in an electrical corridor, you may be interested in determining which points correspond to conductors, which to towers, which to ground and which to vegetation. This can be useful for example to generate alerts when an electric conductor is too close to the vegetation. We all remember the tragic 2018 camp fire in California.

Instead of doing classification manually, which is a painful and time-consuming task, automatic algorithms can do the job for you in an accurate, fast and reproducible way. Terra3D is specialized in developing this kind of automatic tools.

Raw point cloud
Classified point cloud

Example of automatic classification in a railway environment. On the right: raw point cloud in white color. On the left: classified point cloud where each color represents an object class.

Raw point cloud
Classified point cloud

Example of automatic classification in a powerline environment. On the right: raw point cloud in white color. On the left: classified point cloud where each color represents an object class.

What is 3D modeling ?

Modeling is the process of generating a geometrical model from a 3D point cloud. It is usually carried from a previously classified point cloud. For example, in a powerline corridor, wires conductors can be modelled as 3D polylines generated from a set of 3D points classified as such. Those polylines can be obtained by fitting a catenary shape. When geometric models are exported in a common CAD format file such as DWG, DXF or SHP, this process is often called vectorisation. Such models can be easily imported into classical CAD and GIS software.

Example of automatic modeling in a powerline environment. Red polylines correspond to 3D catenary models of wire conductors.
Example of automatic modeling in a railway environment. Red polylines correspond to 3D models of cables while blue polylines correspond to 3D models of rails.

Additional resources

Recent Posts

From Paris to Fontainebleau

At Terra3D we love ping pong. But more than that we really (really) love cycling. Besides, we also have two sites in Parisian region : one at Paris, in the heart of the city of love, and another in Fontainebleau, a city renowned for the forest of Fontainebleau and for the historic Chateau de Fontainebleau, just […]

Read More
Introduction to 3D point clouds

What is a 3D Point Cloud ? A 3D point cloud is a list of cartesian coordinates X, Y and Z representing the geometry of a scene.  Generally composed of millions or billions of points, they are accurate digital representations of the real world. These coordinates correspond to the actual size of objects. Thus, we […]

Read More
A convolution operator for point clouds

Deep learning has revolutionized computer vision and Convolutional Neural Networks (CNN) now produce some great results on tasks like semantic classification and object segmentation for 2D images. Convolution is the fundamental building block of a neural network for image processing, but its definition is not directly adapted to point cloud processing. In this article, we […]

Read More
linkedin facebook pinterest youtube rss twitter instagram facebook-blank rss-blank linkedin-blank pinterest youtube twitter instagram