Armaan Sengupta

Projects / LiDAR Point Clouds

LiDAR Point Clouds

Year 1 · Software · 2023

Wrote filtering, voxel downsampling and a fast furthest-pair search for millions of points.

Processing a large urban LiDAR scan with laspy, Open3D and NumPy: voxel downsampling, a nearest-neighbour outlier filter for dust and artifacts, and a custom furthest-pair search fast enough for 1M+ points.

The project

I wrote a pipeline to manipulate, clean and analyze a massive LiDAR point cloud: removing anomalies, extracting buildings, and finding the two furthest points in a set of millions, a good excuse to try some fun algorithms.

Media

Point cloud coloured by height
Point cloud coloured by height
Top-down view
Top-down view
Height colour map
Height colour map
Reference point marked for a later task
Reference point marked for a later task
Statistical outlier removal vs normal
Statistical outlier removal vs normal
Raw cloud
Raw cloud
Raw cloud at ground level
Raw cloud at ground level
Before and after comparison
Before and after comparison
Cleaned cloud, coloured by height
Cleaned cloud, coloured by height
Extracted buildings
Extracted buildings
Segmented cloud
Segmented cloud
Furthest pair of points found
Furthest pair of points found
Clusters
Clusters
Dense greyscale surface
Dense greyscale surface
Buildings only
Buildings only

See it on the interactive siteCode and write-up ↗Code & details (GitHub) ↗3D point cloud online ↗